tutorials/TrainData_GSE135167_mouse_eye_genelist.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e48c5f86-f0aa-4963-8913-604443fa0a3c",
   "metadata": {},
   "source": [
    "# Tutorial: Train single cell data using genelist method"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5aac073f-ef61-49f9-bf3e-13b2a3137121",
   "metadata": {},
   "source": [
    "## Introduction\n",
    "\n",
    "This tutorial demonstrates how to use **scGenAI** to train a model for predicting cell types from single-cell RNA sequencing (scRNA-seq) data. In this example, we apply the **gene list** method to generate input sequences.\n",
    "\n",
    "**Note:**\n",
    "1. The dataset used for this tutorial is reported in [**Lehmann et. al., 2020**](https://pubmed.ncbi.nlm.nih.gov/32196081/). The input data train and validation files can be downloaded from link of [**GSE135167_mouse_eye_genelist**](https://zenodo.org/records/14036352).\n",
    "2. In this tutorial, we will load the model template directly from the cache folder.\n",
    "3. While the model in this example is trained to predict cell types, it can be used to predict any feature in practical applications.\n",
    "4. In this tutorial, the gene list is built by extracting the top 1000 variable genes from the data, in the practical case, this can be any list that may facilitate the prediction.\n",
    "\n",
    "**Reference**\n",
    "1. Lehmann, G. L., Hanke-Gogokhia, C., Hu, Y., Bareja, R., Salfati, Z., Ginsberg, M., Nolan, D. J., Mendez-Huergo, S. P., Dalotto-Moreno, T., Wojcinski, A., Ochoa, F., Zeng, S., Cerliani, J. P., Panagis, L., Zager, P. J., Mullins, R. F., Ogura, S., Lutty, G. A., Bang, J., Zippin, J. H., … Benedicto, I. (2020). Single-cell profiling reveals an endothelium-mediated immunomodulatory pathway in the eye choroid. The Journal of experimental medicine, 217(6), e20190730. https://doi.org/10.1084/jem.20190730\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8294154a-4036-4788-b358-0e91342ac224",
   "metadata": {},
   "source": [
    "## Train the Model Using Genelist (Multiple GPU)\n",
    "\n",
    "The training process can be easily initiated by calling the `run_training_from_config` function in **scGenAI**, using the input files specified in a configuration YAML file. \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "2989a022-6bea-4f9c-861d-452b0d85e11f",
   "metadata": {},
   "outputs": [],
   "source": [
    "from scGenAI.training.train import run_training_from_config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "5a140f4c-6d3c-4a18-9f15-87d148179ac8",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "model_dir: /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/\n",
      "train_file: /home/ubuntu/fulldataset/GSE135167_mouse_eye.train.h5ad\n",
      "val_file: /home/ubuntu/fulldataset/GSE135167_mouse_eye.val.h5ad\n",
      "glstfile: /home/ubuntu/fulldataset/GSE135167_mouse_eye.highly_variable_genes_TOP1000.txt\n",
      "log_dir: /home/ubuntu/fulldataset/logs/\n",
      "output_dir: /home/ubuntu/scgenai/examples/output\n",
      "seed: 1314521\n",
      "cache_dir: /home/ubuntu/scgenai/examples/tmp/cache\n",
      "mode: Train\n",
      "min_cells: 50\n",
      "target_feature: ct\n",
      "multiomics: No\n",
      "keepIntermediateFiles: Yes\n",
      "savelog: Yes\n",
      "savesetting: Yes\n",
      "model_backbone_name: llama\n",
      "model_backbone_size: small\n",
      "context_method: genelist\n",
      "num_bins: 50\n",
      "optimizer: AdamW\n",
      "weight_decay: 0.01\n",
      "depth: 2\n",
      "max_length: 2048\n",
      "batch_size: 1\n",
      "learning_rate: 1e-05\n",
      "num_epochs: 30\n",
      "world_size: 1\n",
      "evaluate_during_training: True\n",
      "hidden_size: 256\n",
      "num_layers: 6\n",
      "num_heads: 4\n",
      "intermediate_size: 1024\n",
      "stride: 1024\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-10-21 17:23:38.349 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 0, Total GPUs used: 4\n",
      "2024-10-21 17:23:38.355 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 3, Total GPUs used: 4\n",
      "2024-10-21 17:23:38.364 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 2, Total GPUs used: 4\n",
      "2024-10-21 17:23:38.365 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 1, Total GPUs used: 4\n",
      "2024-10-21 17:23:40.556 | INFO     | scGenAI.training.train:model_train_and_eval:214 - Total number of trainable parameters: 72371207; Max Sequence Legth: 1024 genes\n",
      "Epoch 1/30: 100%|██████████| 1513/1513 [07:10<00:00,  3.51it/s, GPU_ID=0, train_accuracy=0.862, train_loss=0.451]\n",
      "Epoch 1/30: 100%|██████████| 1513/1513 [07:10<00:00,  3.51it/s, GPU_ID=3, train_accuracy=0.873, train_loss=0.409]\n",
      "Epoch 1/30: 100%|██████████| 1513/1513 [07:10<00:00,  3.51it/s, GPU_ID=2, train_accuracy=0.857, train_loss=0.435]\n",
      "Epoch 1/30: 100%|██████████| 1513/1513 [07:10<00:00,  3.51it/s, GPU_ID=1, train_accuracy=0.853, train_loss=0.455]\n",
      "Evaluating Epoch 1/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.15it/s]\n",
      "Evaluating Epoch 1/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.09it/s]\n",
      "Evaluating Epoch 1/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.83it/s]\n",
      "2024-10-21 17:32:21.953 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 1/30, Train Loss: 0.4506, Train Accuracy: 0.8619, Context-level Val Loss: 0.1340, Context-level Val Accuracy: 0.9564\n",
      "Evaluating Epoch 1/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.79it/s]\n",
      "Epoch 2/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 17:32:24.671 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 17:32:24.821 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 17:32:24.826 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 17:32:24.827 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 2/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.978, train_loss=0.0776]\n",
      "\n",
      "Epoch 2/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.971, train_loss=0.0886]\n",
      "Epoch 2/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.971, train_loss=0.102]\n",
      "Evaluating Epoch 2/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.15it/s]\n",
      "Evaluating Epoch 2/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.06it/s]\n",
      "Evaluating Epoch 2/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.84it/s]\n",
      "2024-10-21 17:41:06.494 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 2/30, Train Loss: 0.0886, Train Accuracy: 0.9709, Context-level Val Loss: 0.1274, Context-level Val Accuracy: 0.9617\n",
      "Evaluating Epoch 2/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.80it/s]\n",
      "Epoch 3/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 17:41:09.252 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 17:41:09.403 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 17:41:09.408 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 17:41:09.409 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 3/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.981, train_loss=0.0762]\n",
      "\n",
      "Epoch 3/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.983, train_loss=0.0517]\n",
      "Epoch 3/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.98, train_loss=0.0725]\n",
      "Evaluating Epoch 3/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.17it/s]\n",
      "Evaluating Epoch 3/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.09it/s]\n",
      "Evaluating Epoch 3/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.91it/s]\n",
      "2024-10-21 17:49:50.450 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 3/30, Train Loss: 0.0501, Train Accuracy: 0.9861, Context-level Val Loss: 0.0802, Context-level Val Accuracy: 0.9755\n",
      "Evaluating Epoch 3/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.85it/s]\n",
      "Epoch 4/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 17:49:53.182 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 17:49:53.332 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 17:49:53.335 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 17:49:53.336 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 4/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.987, train_loss=0.0469]\n",
      "\n",
      "Epoch 4/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.51it/s, GPU_ID=0, train_accuracy=0.99, train_loss=0.0349]\n",
      "Epoch 4/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.984, train_loss=0.0608]\n",
      "Evaluating Epoch 4/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.18it/s]\n",
      "Evaluating Epoch 4/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.10it/s]\n",
      "Evaluating Epoch 4/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.93it/s]\n",
      "2024-10-21 17:58:34.232 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 4/30, Train Loss: 0.0349, Train Accuracy: 0.9901, Context-level Val Loss: 0.0628, Context-level Val Accuracy: 0.9861\n",
      "Evaluating Epoch 4/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.81it/s]\n",
      "Epoch 5/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 17:58:36.966 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 17:58:37.113 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 17:58:37.117 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 17:58:37.118 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 5/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.51it/s, GPU_ID=0, train_accuracy=0.989, train_loss=0.0286]\n",
      "\n",
      "Epoch 5/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.984, train_loss=0.0532]\n",
      "Epoch 5/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.985, train_loss=0.0426]\n",
      "Evaluating Epoch 5/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.17it/s]\n",
      "Evaluating Epoch 5/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 5/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.92it/s]\n",
      "2024-10-21 18:07:18.001 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 5/30, Train Loss: 0.0286, Train Accuracy: 0.9894, Context-level Val Loss: 0.0713, Context-level Val Accuracy: 0.9769\n",
      "Evaluating Epoch 5/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.78it/s]\n",
      "Epoch 6/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=3, train_accuracy=0.992, train_loss=0.0283] \n",
      "Epoch 6/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.993, train_loss=0.0255]\n",
      "Epoch 6/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=2, train_accuracy=0.989, train_loss=0.0385]\n",
      "Epoch 6/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.985, train_loss=0.0487]\n",
      "Evaluating Epoch 6/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 6/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.10it/s]\n",
      "Evaluating Epoch 6/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.88it/s]\n",
      "2024-10-21 18:16:01.354 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 6/30, Train Loss: 0.0255, Train Accuracy: 0.9927, Context-level Val Loss: 0.0657, Context-level Val Accuracy: 0.9835\n",
      "Evaluating Epoch 6/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.77it/s]\n",
      "Epoch 7/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=3, train_accuracy=0.993, train_loss=0.0274]\n",
      "Epoch 7/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.995, train_loss=0.0148]\n",
      "Epoch 7/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=2, train_accuracy=0.99, train_loss=0.0325]\n",
      "Epoch 7/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.991, train_loss=0.0273]\n",
      "Evaluating Epoch 7/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.16it/s]\n",
      "Evaluating Epoch 7/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.04it/s]\n",
      "Evaluating Epoch 7/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.87it/s]\n",
      "2024-10-21 18:24:44.508 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 7/30, Train Loss: 0.0148, Train Accuracy: 0.9954, Context-level Val Loss: 0.0816, Context-level Val Accuracy: 0.9822\n",
      "Evaluating Epoch 7/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.73it/s]\n",
      "Epoch 8/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=3, train_accuracy=0.997, train_loss=0.0104] \n",
      "Epoch 8/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.99, train_loss=0.0388]\n",
      "Epoch 8/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.993, train_loss=0.0192]\n",
      "Epoch 8/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.993, train_loss=0.0199]\n",
      "Evaluating Epoch 8/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 8/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.04it/s]\n",
      "Evaluating Epoch 8/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.83it/s]\n",
      "2024-10-21 18:33:28.288 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 8/30, Train Loss: 0.0192, Train Accuracy: 0.9927, Context-level Val Loss: 0.0966, Context-level Val Accuracy: 0.9802\n",
      "Evaluating Epoch 8/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.78it/s]\n",
      "Epoch 9/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=3, train_accuracy=0.991, train_loss=0.0221] \n",
      "Epoch 9/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.995, train_loss=0.0137]\n",
      "Epoch 9/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=2, train_accuracy=0.996, train_loss=0.0137]\n",
      "Epoch 9/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.991, train_loss=0.021]\n",
      "Evaluating Epoch 9/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 9/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.05it/s]\n",
      "Evaluating Epoch 9/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.85it/s]\n",
      "2024-10-21 18:42:11.428 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 9/30, Train Loss: 0.0137, Train Accuracy: 0.9954, Context-level Val Loss: 0.0828, Context-level Val Accuracy: 0.9868\n",
      "Evaluating Epoch 9/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.71it/s]\n",
      "Epoch 10/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 18:42:14.148 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 18:42:14.306 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 18:42:14.311 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 18:42:14.312 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 10/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.997, train_loss=0.0169] \n",
      "Epoch 10/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.997, train_loss=0.0127]\n",
      "Epoch 10/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.996, train_loss=0.0145]\n",
      "Epoch 10/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.994, train_loss=0.0234]\n",
      "Evaluating Epoch 10/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.14it/s]\n",
      "Evaluating Epoch 10/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.04it/s]\n",
      "Evaluating Epoch 10/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.85it/s]\n",
      "Evaluating Epoch 10/30: 100%|█████████▉| 1506/1513 [01:29<00:00, 16.52it/s]2024-10-21 18:50:55.871 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 10/30, Train Loss: 0.0145, Train Accuracy: 0.9960, Context-level Val Loss: 0.0891, Context-level Val Accuracy: 0.9828\n",
      "Evaluating Epoch 10/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.77it/s]\n",
      "Epoch 11/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=3, train_accuracy=0.993, train_loss=0.0244] \n",
      "Epoch 11/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=2, train_accuracy=0.993, train_loss=0.0297]\n",
      "Epoch 11/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.997, train_loss=0.0172]\n",
      "Epoch 11/30: 100%|██████████| 1513/1513 [07:11<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.997, train_loss=0.0125]\n",
      "Evaluating Epoch 11/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 11/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.02it/s]\n",
      "Evaluating Epoch 11/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.83it/s]\n",
      "2024-10-21 18:59:39.063 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 11/30, Train Loss: 0.0172, Train Accuracy: 0.9967, Context-level Val Loss: 0.0669, Context-level Val Accuracy: 0.9868\n",
      "Evaluating Epoch 11/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.75it/s]\n",
      "Epoch 12/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 18:59:41.794 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 18:59:41.944 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 18:59:41.948 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 18:59:41.949 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 12/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.996, train_loss=0.0138]]\n",
      "\n",
      "Epoch 12/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.997, train_loss=0.00819]\n",
      "Epoch 12/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.997, train_loss=0.00892]\n",
      "Evaluating Epoch 12/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 12/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.02it/s]\n",
      "Evaluating Epoch 12/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.81it/s]\n",
      "2024-10-21 19:08:24.033 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 12/30, Train Loss: 0.0082, Train Accuracy: 0.9974, Context-level Val Loss: 0.0818, Context-level Val Accuracy: 0.9874\n",
      "Evaluating Epoch 12/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.69it/s]\n",
      "Epoch 13/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 19:08:26.732 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 19:08:26.882 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 19:08:26.887 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 19:08:26.888 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 13/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=2, train_accuracy=0.997, train_loss=0.0103]]\n",
      "\n",
      "Epoch 13/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.999, train_loss=0.00952]\n",
      "Epoch 13/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.994, train_loss=0.0198]\n",
      "Evaluating Epoch 13/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.08it/s]\n",
      "Evaluating Epoch 13/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.96it/s]\n",
      "Evaluating Epoch 13/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.76it/s]\n",
      "2024-10-21 19:17:09.580 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 13/30, Train Loss: 0.0095, Train Accuracy: 0.9987, Context-level Val Loss: 0.0834, Context-level Val Accuracy: 0.9868\n",
      "Evaluating Epoch 13/30: 100%|██████████| 1513/1513 [01:31<00:00, 16.62it/s]\n",
      "Epoch 14/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.999, train_loss=0.00676]\n",
      "Epoch 14/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=0, train_accuracy=0.995, train_loss=0.0224]]\n",
      "\n",
      "Epoch 14/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.999, train_loss=0.00671]\n",
      "Evaluating Epoch 14/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 14/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.06it/s]\n",
      "Evaluating Epoch 14/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.83it/s]\n",
      "2024-10-21 19:25:53.627 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 14/30, Train Loss: 0.0224, Train Accuracy: 0.9954, Context-level Val Loss: 0.0609, Context-level Val Accuracy: 0.9861\n",
      "Evaluating Epoch 14/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.65it/s]\n",
      "Epoch 15/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.999, train_loss=0.00119] \n",
      "Epoch 15/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=0, train_accuracy=0.995, train_loss=0.0168]\n",
      "Epoch 15/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.996, train_loss=0.0203]\n",
      "Epoch 15/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.998, train_loss=0.00499]\n",
      "Evaluating Epoch 15/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 15/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.00it/s]\n",
      "Evaluating Epoch 15/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.78it/s]\n",
      "2024-10-21 19:34:38.043 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 15/30, Train Loss: 0.0168, Train Accuracy: 0.9954, Context-level Val Loss: 0.0645, Context-level Val Accuracy: 0.9901\n",
      "Evaluating Epoch 15/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.68it/s]\n",
      "Epoch 16/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 19:34:40.768 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 19:34:40.928 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 19:34:40.932 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 19:34:40.933 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 16/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=3, train_accuracy=1, train_loss=0.000245]47]\n",
      "Epoch 16/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=2, train_accuracy=0.999, train_loss=0.00225]\n",
      "\n",
      "Epoch 16/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.999, train_loss=0.00447]\n",
      "Evaluating Epoch 16/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.15it/s]\n",
      "Evaluating Epoch 16/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.02it/s]\n",
      "Evaluating Epoch 16/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.81it/s]\n",
      "2024-10-21 19:43:23.163 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 16/30, Train Loss: 0.0048, Train Accuracy: 0.9993, Context-level Val Loss: 0.0770, Context-level Val Accuracy: 0.9888\n",
      "Evaluating Epoch 16/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.71it/s]\n",
      "Epoch 17/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=3, train_accuracy=1, train_loss=0.000571]]8]\n",
      "Epoch 17/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=1, train_loss=0.00057]388]\n",
      "\n",
      "Epoch 17/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.997, train_loss=0.011]\n",
      "Evaluating Epoch 17/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 17/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.06it/s]\n",
      "Evaluating Epoch 17/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.83it/s]\n",
      "2024-10-21 19:52:06.786 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 17/30, Train Loss: 0.0006, Train Accuracy: 1.0000, Context-level Val Loss: 0.0681, Context-level Val Accuracy: 0.9894\n",
      "Evaluating Epoch 17/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.69it/s]\n",
      "Epoch 18/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.49it/s, GPU_ID=0, train_accuracy=1, train_loss=0.000128]5]\n",
      "Epoch 18/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=1, train_loss=6.83e-5]\n",
      "\n",
      "Epoch 18/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.999, train_loss=0.0125]\n",
      "Evaluating Epoch 18/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.14it/s]\n",
      "Evaluating Epoch 18/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.03it/s]\n",
      "Evaluating Epoch 18/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.80it/s]\n",
      "2024-10-21 20:00:50.902 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 18/30, Train Loss: 0.0001, Train Accuracy: 1.0000, Context-level Val Loss: 0.0664, Context-level Val Accuracy: 0.9907\n",
      "Evaluating Epoch 18/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.67it/s]\n",
      "Epoch 19/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 20:00:53.634 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 20:00:53.782 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 20:00:53.787 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 20:00:53.788 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 19/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.994, train_loss=0.0217] \n",
      "Epoch 19/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=0.997, train_loss=0.0141]\n",
      "Epoch 19/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=2, train_accuracy=0.995, train_loss=0.0181]\n",
      "Epoch 19/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=0.994, train_loss=0.0205]\n",
      "Evaluating Epoch 19/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.14it/s]\n",
      "Evaluating Epoch 19/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.02it/s]\n",
      "Evaluating Epoch 19/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.81it/s]\n",
      "2024-10-21 20:09:36.013 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 19/30, Train Loss: 0.0141, Train Accuracy: 0.9974, Context-level Val Loss: 0.0991, Context-level Val Accuracy: 0.9855\n",
      "Evaluating Epoch 19/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.70it/s]\n",
      "Epoch 20/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=0.999, train_loss=0.00647]\n",
      "Epoch 20/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=0.997, train_loss=0.0121]]\n",
      "\n",
      "Epoch 20/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=0.997, train_loss=0.0133]\n",
      "Evaluating Epoch 20/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 20/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.03it/s]\n",
      "Evaluating Epoch 20/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.79it/s]\n",
      "2024-10-21 20:18:20.256 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 20/30, Train Loss: 0.0058, Train Accuracy: 0.9987, Context-level Val Loss: 0.0490, Context-level Val Accuracy: 0.9907\n",
      "Evaluating Epoch 20/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.68it/s]\n",
      "Epoch 21/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 20:18:22.978 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 20:18:23.136 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 20:18:23.141 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 20:18:23.141 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 21/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=1, train_loss=0.000123]\n",
      "\n",
      "Epoch 21/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=3, train_accuracy=1, train_loss=0.000289]\n",
      "Epoch 21/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=1, train_loss=0.000246]\n",
      "Evaluating Epoch 21/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.11it/s]\n",
      "Evaluating Epoch 21/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.01it/s]\n",
      "Evaluating Epoch 21/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.79it/s]\n",
      "2024-10-21 20:27:05.774 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 21/30, Train Loss: 0.0001, Train Accuracy: 1.0000, Context-level Val Loss: 0.0488, Context-level Val Accuracy: 0.9914\n",
      "Evaluating Epoch 21/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.65it/s]\n",
      "Epoch 22/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 20:27:08.509 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 20:27:08.657 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 20:27:08.661 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 20:27:08.662 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 22/30: 100%|██████████| 1513/1513 [07:16<00:00,  3.47it/s, GPU_ID=2, train_accuracy=1, train_loss=6.84e-5]]\n",
      "\n",
      "Epoch 22/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=1, train_loss=0.000158]\n",
      "Epoch 22/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=1, train_loss=6.5e-5]\n",
      "Evaluating Epoch 22/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 22/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.01it/s]\n",
      "Evaluating Epoch 22/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.79it/s]\n",
      "2024-10-21 20:35:51.370 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 22/30, Train Loss: 0.0002, Train Accuracy: 1.0000, Context-level Val Loss: 0.0548, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 22/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.65it/s]\n",
      "Epoch 23/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 20:35:54.101 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 20:35:54.261 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 20:35:54.265 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 20:35:54.266 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 23/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=1, train_loss=3.32e-5]\n",
      "Epoch 23/30: 100%|██████████| 1513/1513 [07:16<00:00,  3.47it/s, GPU_ID=2, train_accuracy=1, train_loss=2.8e-5]\n",
      "Epoch 23/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=3, train_accuracy=1, train_loss=2.61e-5]\n",
      "Epoch 23/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=1, train_loss=2.58e-5]\n",
      "Evaluating Epoch 23/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 23/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.03it/s]\n",
      "Evaluating Epoch 23/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.79it/s]\n",
      "2024-10-21 20:44:36.799 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 23/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0566, Context-level Val Accuracy: 0.9934\n",
      "Evaluating Epoch 23/30: 100%|██████████| 1513/1513 [01:31<00:00, 16.61it/s]\n",
      "Epoch 24/30:   0%|          | 0/1513 [00:00<?, ?it/s]2024-10-21 20:44:39.530 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/scGenAI_model.pt\n",
      "2024-10-21 20:44:39.679 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "2024-10-21 20:44:39.684 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 20:44:39.685 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model\n",
      "Epoch 24/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=3, train_accuracy=1, train_loss=1.51e-5]\n",
      "Epoch 24/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.47it/s, GPU_ID=2, train_accuracy=1, train_loss=1.59e-5]\n",
      "Epoch 24/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=1, train_loss=1.77e-5]\n",
      "Epoch 24/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=1, train_accuracy=1, train_loss=1.44e-5]\n",
      "Evaluating Epoch 24/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 24/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.02it/s]\n",
      "Evaluating Epoch 24/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.80it/s]\n",
      "2024-10-21 20:53:22.028 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 24/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0580, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 24/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.69it/s]\n",
      "Epoch 25/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.49it/s, GPU_ID=3, train_accuracy=1, train_loss=8.95e-6]\n",
      "Epoch 25/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=0, train_accuracy=1, train_loss=9.71e-6]\n",
      "\n",
      "Epoch 25/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=1, train_loss=8.38e-6]\n",
      "Evaluating Epoch 25/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 25/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.03it/s]\n",
      "Evaluating Epoch 25/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.82it/s]\n",
      "2024-10-21 21:02:05.848 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 25/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0596, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 25/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.65it/s]\n",
      "Epoch 26/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=1, train_loss=5.08e-6]\n",
      "Epoch 26/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=1, train_loss=5.24e-6]\n",
      "\n",
      "Epoch 26/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=1, train_loss=4.82e-6]\n",
      "Evaluating Epoch 26/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 26/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.07it/s]\n",
      "Evaluating Epoch 26/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.83it/s]\n",
      "2024-10-21 21:10:50.023 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 26/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0616, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 26/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.67it/s]\n",
      "Epoch 27/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=1, train_loss=2.75e-6]\n",
      "Epoch 27/30: 100%|██████████| 1513/1513 [07:13<00:00,  3.49it/s, GPU_ID=0, train_accuracy=1, train_loss=2.8e-6]\n",
      "Epoch 27/30: 100%|██████████| 1513/1513 [07:15<00:00,  3.48it/s, GPU_ID=2, train_accuracy=1, train_loss=2.82e-6]\n",
      "Epoch 27/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=1, train_loss=2.67e-6]\n",
      "Evaluating Epoch 27/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 27/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.01it/s]\n",
      "Evaluating Epoch 27/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.80it/s]\n",
      "2024-10-21 21:19:34.506 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 27/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0640, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 27/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.70it/s]\n",
      "Epoch 28/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.49it/s, GPU_ID=3, train_accuracy=1, train_loss=1.41e-6]\n",
      "Epoch 28/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=1, train_loss=1.46e-6]\n",
      "\n",
      "Epoch 28/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=1, train_loss=1.42e-6]\n",
      "Evaluating Epoch 28/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.11it/s]\n",
      "Evaluating Epoch 28/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.05it/s]\n",
      "Evaluating Epoch 28/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.80it/s]\n",
      "2024-10-21 21:28:18.515 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 28/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0668, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 28/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.71it/s]\n",
      "Epoch 29/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=1, train_loss=7.41e-7]\n",
      "Epoch 29/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=3, train_accuracy=1, train_loss=7.16e-7]\n",
      "Epoch 29/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.49it/s, GPU_ID=0, train_accuracy=1, train_loss=7.39e-7]\n",
      "Epoch 29/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=1, train_loss=7.37e-7]\n",
      "Evaluating Epoch 29/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.12it/s]\n",
      "Evaluating Epoch 29/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.04it/s]\n",
      "Evaluating Epoch 29/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.80it/s]\n",
      "2024-10-21 21:37:02.592 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 29/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0699, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 29/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.70it/s]\n",
      "Epoch 30/30: 100%|██████████| 1513/1513 [07:14<00:00,  3.48it/s, GPU_ID=2, train_accuracy=1, train_loss=3.71e-7]\n",
      "\n",
      "\n",
      "Epoch 30/30: 100%|██████████| 1513/1513 [07:12<00:00,  3.50it/s, GPU_ID=1, train_accuracy=1, train_loss=3.76e-7]\n",
      "Evaluating Epoch 30/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.13it/s]\n",
      "Evaluating Epoch 30/30: 100%|██████████| 1513/1513 [01:28<00:00, 17.06it/s]\n",
      "Evaluating Epoch 30/30: 100%|██████████| 1513/1513 [01:29<00:00, 16.82it/s]\n",
      "2024-10-21 21:45:46.531 | INFO     | scGenAI.training.train:evaluate_and_save_predictions:328 - Epoch 30/30, Train Loss: 0.0000, Train Accuracy: 1.0000, Context-level Val Loss: 0.0731, Context-level Val Accuracy: 0.9927\n",
      "Evaluating Epoch 30/30: 100%|██████████| 1513/1513 [01:30<00:00, 16.70it/s]\n",
      "2024-10-21 21:45:48.093 | INFO     | scGenAI.utils.distributed:savemodel:54 - Model state saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/last_model/scGenAI_model.pt\n",
      "2024-10-21 21:45:48.230 | INFO     | scGenAI.utils.distributed:savemodel:59 - Tokenizer saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/last_model\n",
      "2024-10-21 21:45:48.231 | INFO     | scGenAI.utils.distributed:savemodel:75 - Custom vocab and label encoder saved\n",
      "2024-10-21 21:45:48.232 | INFO     | scGenAI.utils.distributed:savemodel:81 - Model configuration saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/last_model\n",
      "2024-10-21 21:45:48.264 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "2024-10-21 21:45:48.264 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "2024-10-21 21:45:48.264 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "2024-10-21 21:45:48.264 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "\u001b[32m2024-10-21 21:45:50.312\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 1\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:52.404\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 1 - Validation Accuracy: 0.9551\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:52.517\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 2\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:54.615\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 2 - Validation Accuracy: 0.9564\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:54.728\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 3\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:56.820\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 3 - Validation Accuracy: 0.9676\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:56.933\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 4\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:59.026\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 4 - Validation Accuracy: 0.9788\u001b[0m\n",
      "\u001b[32m2024-10-21 21:45:59.139\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 5\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:01.223\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 5 - Validation Accuracy: 0.9722\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:01.335\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 6\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:03.406\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 6 - Validation Accuracy: 0.9841\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:03.519\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 7\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:05.578\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 7 - Validation Accuracy: 0.9815\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:05.691\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 8\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:07.736\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 8 - Validation Accuracy: 0.9762\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:07.848\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 9\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:10.058\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 9 - Validation Accuracy: 0.9835\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:10.169\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 10\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:12.191\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 10 - Validation Accuracy: 0.9802\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:12.302\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 11\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:14.310\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 11 - Validation Accuracy: 0.9861\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:14.421\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 12\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:16.417\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 12 - Validation Accuracy: 0.9868\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:16.528\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 13\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:18.511\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 13 - Validation Accuracy: 0.9861\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:18.622\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 14\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:20.676\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 14 - Validation Accuracy: 0.9868\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:20.786\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 15\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:22.852\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 15 - Validation Accuracy: 0.9901\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:22.962\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 16\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:25.027\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 16 - Validation Accuracy: 0.9874\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:25.137\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 17\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:27.199\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 17 - Validation Accuracy: 0.9888\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:27.309\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 18\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:29.366\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 18 - Validation Accuracy: 0.9888\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:29.475\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 19\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:31.519\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 19 - Validation Accuracy: 0.9835\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:31.628\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 20\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:33.658\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 20 - Validation Accuracy: 0.9894\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:33.768\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 21\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:35.791\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 21 - Validation Accuracy: 0.9907\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:35.900\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 22\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:37.904\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 22 - Validation Accuracy: 0.9921\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:38.013\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 23\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:40.005\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 23 - Validation Accuracy: 0.9921\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:40.114\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 24\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:42.092\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 24 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:42.201\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 25\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:44.167\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 25 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:44.275\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 26\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:46.314\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 26 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:46.423\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 27\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:48.472\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 27 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:48.580\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 28\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:50.849\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 28 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:50.957\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 29\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:53.003\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 29 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:53.111\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mmerge_predictions_across_gpus\u001b[0m:\u001b[36m147\u001b[0m - \u001b[1mMerging predictions for epoch 30\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:55.156\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mpostprocess_epoch\u001b[0m:\u001b[36m211\u001b[0m - \u001b[1mEpoch 30 - Validation Accuracy: 0.9927\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:55.266\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36msummarize_epochs\u001b[0m:\u001b[36m234\u001b[0m - \u001b[1mCompleted post-processing and summary of all epochs.\u001b[0m\n",
      "\u001b[32m2024-10-21 21:46:55.269\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mcombine_epoch_results\u001b[0m:\u001b[36m61\u001b[0m - \u001b[1mCombined results saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/combined_epoch_results.csv.\u001b[0m\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/train_summary.pdf\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "config_file = '/home/ubuntu/scgenai/tutorials/yaml/config_genelist_GSE135167_mouse_eye.yaml'\n",
    "run_training_from_config(config_file)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2fe550d5-16c0-425a-bfd7-d257c1ecfe1d",
   "metadata": {},
   "source": [
    "## Post-Training Model and Results\n",
    "\n",
    "Upon completion of the training, the trained model will be saved in the `model_dir` as defined in the configuration file. Within this directory, two subfolders will be created: `best_model` and `last_model`. The `best_model` folder contains the model that achieved the highest training performance, while the `last_model` contains the model from the final epoch. Additionally, a `combined_epoch_results.csv` file will be generated in the `model_dir`, summarizing the training performance metrics across all epochs."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cc88e128-eaba-4557-8fca-eae0966e482d",
   "metadata": {},
   "source": [
    "### Training Summary\n",
    "\n",
    "In addition to the output summary files generated after training, the training summary can be directly accessed in Python by loading the saved configuration file.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "9cd73cce-7ddc-47ab-b59f-fc4ac955b57f",
   "metadata": {},
   "outputs": [],
   "source": [
    "from scGenAI.config import Config\n",
    "from scGenAI.data.postprocess import TrainSummary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "9bc20420-8143-4ff7-b474-d4a3f6936991",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Epoch</th>\n",
       "      <th>Validation Accuracy</th>\n",
       "      <th>Train Loss</th>\n",
       "      <th>Train Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>0.955056</td>\n",
       "      <td>4.505679e-01</td>\n",
       "      <td>0.861864</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>0.956378</td>\n",
       "      <td>8.856105e-02</td>\n",
       "      <td>0.970919</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>0.967614</td>\n",
       "      <td>5.010466e-02</td>\n",
       "      <td>0.986120</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0.978850</td>\n",
       "      <td>3.490974e-02</td>\n",
       "      <td>0.990086</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>0.972241</td>\n",
       "      <td>2.856238e-02</td>\n",
       "      <td>0.989425</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>6</td>\n",
       "      <td>0.984137</td>\n",
       "      <td>2.545965e-02</td>\n",
       "      <td>0.992730</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>7</td>\n",
       "      <td>0.981494</td>\n",
       "      <td>1.477643e-02</td>\n",
       "      <td>0.995373</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>8</td>\n",
       "      <td>0.976206</td>\n",
       "      <td>1.924984e-02</td>\n",
       "      <td>0.992730</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>9</td>\n",
       "      <td>0.983477</td>\n",
       "      <td>1.367687e-02</td>\n",
       "      <td>0.995373</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>10</td>\n",
       "      <td>0.980172</td>\n",
       "      <td>1.445226e-02</td>\n",
       "      <td>0.996034</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>11</td>\n",
       "      <td>0.986120</td>\n",
       "      <td>1.719473e-02</td>\n",
       "      <td>0.996695</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>12</td>\n",
       "      <td>0.986781</td>\n",
       "      <td>8.187722e-03</td>\n",
       "      <td>0.997356</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>13</td>\n",
       "      <td>0.986120</td>\n",
       "      <td>9.515192e-03</td>\n",
       "      <td>0.998678</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>14</td>\n",
       "      <td>0.986781</td>\n",
       "      <td>2.240826e-02</td>\n",
       "      <td>0.995373</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>15</td>\n",
       "      <td>0.990086</td>\n",
       "      <td>1.679708e-02</td>\n",
       "      <td>0.995373</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>16</td>\n",
       "      <td>0.987442</td>\n",
       "      <td>4.841243e-03</td>\n",
       "      <td>0.999339</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>17</td>\n",
       "      <td>0.988764</td>\n",
       "      <td>5.704443e-04</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>18</td>\n",
       "      <td>0.988764</td>\n",
       "      <td>1.283778e-04</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>19</td>\n",
       "      <td>0.983477</td>\n",
       "      <td>1.407412e-02</td>\n",
       "      <td>0.997356</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>20</td>\n",
       "      <td>0.989425</td>\n",
       "      <td>5.786440e-03</td>\n",
       "      <td>0.998678</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>21</td>\n",
       "      <td>0.990747</td>\n",
       "      <td>1.227286e-04</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>22</td>\n",
       "      <td>0.992069</td>\n",
       "      <td>1.578189e-04</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>23</td>\n",
       "      <td>0.992069</td>\n",
       "      <td>3.316078e-05</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>24</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>1.770001e-05</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>25</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>9.713663e-06</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>26</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>5.290310e-06</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>27</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>2.798589e-06</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>28</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>1.438855e-06</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>29</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>7.387332e-07</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29</th>\n",
       "      <td>30</td>\n",
       "      <td>0.992730</td>\n",
       "      <td>3.640094e-07</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    Epoch  Validation Accuracy    Train Loss  Train Accuracy\n",
       "0       1             0.955056  4.505679e-01        0.861864\n",
       "1       2             0.956378  8.856105e-02        0.970919\n",
       "2       3             0.967614  5.010466e-02        0.986120\n",
       "3       4             0.978850  3.490974e-02        0.990086\n",
       "4       5             0.972241  2.856238e-02        0.989425\n",
       "5       6             0.984137  2.545965e-02        0.992730\n",
       "6       7             0.981494  1.477643e-02        0.995373\n",
       "7       8             0.976206  1.924984e-02        0.992730\n",
       "8       9             0.983477  1.367687e-02        0.995373\n",
       "9      10             0.980172  1.445226e-02        0.996034\n",
       "10     11             0.986120  1.719473e-02        0.996695\n",
       "11     12             0.986781  8.187722e-03        0.997356\n",
       "12     13             0.986120  9.515192e-03        0.998678\n",
       "13     14             0.986781  2.240826e-02        0.995373\n",
       "14     15             0.990086  1.679708e-02        0.995373\n",
       "15     16             0.987442  4.841243e-03        0.999339\n",
       "16     17             0.988764  5.704443e-04        1.000000\n",
       "17     18             0.988764  1.283778e-04        1.000000\n",
       "18     19             0.983477  1.407412e-02        0.997356\n",
       "19     20             0.989425  5.786440e-03        0.998678\n",
       "20     21             0.990747  1.227286e-04        1.000000\n",
       "21     22             0.992069  1.578189e-04        1.000000\n",
       "22     23             0.992069  3.316078e-05        1.000000\n",
       "23     24             0.992730  1.770001e-05        1.000000\n",
       "24     25             0.992730  9.713663e-06        1.000000\n",
       "25     26             0.992730  5.290310e-06        1.000000\n",
       "26     27             0.992730  2.798589e-06        1.000000\n",
       "27     28             0.992730  1.438855e-06        1.000000\n",
       "28     29             0.992730  7.387332e-07        1.000000\n",
       "29     30             0.992730  3.640094e-07        1.000000"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "config = Config(config_file)\n",
    "summary = TrainSummary(config.model_dir)\n",
    "dfsummary = summary.table()\n",
    "dfsummary"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a0ba9299-9ce7-4625-8f21-c49a6e3fd75a",
   "metadata": {},
   "source": [
    "It can also be plotted using the `summary.plot`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "24962a8b-d8b7-4b38-b982-10633416b86e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Plot saved to /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/train_summary.pdf\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "summary.plot(saveplot=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "346b4f66-b17c-462e-947f-0a3d08972cb4",
   "metadata": {},
   "source": [
    "## Prediction Using the Trained Model\n",
    "\n",
    "Similar to the training step, predictions can be run using the `run_prediction_from_config` function in scGenAI with a simple configuration file. Unlike the training configuration, the prediction configuration file only requires the following parameters to be defined: `mode`, `predict_file`, `batch_size`, `model_dir`, and `outputfile`. All other parameters will be adopted from the trained model.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "10cd73c8-8ea4-44cf-a2c5-6b2f4aecd53b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from scGenAI.prediction.predict import run_prediction_from_config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "6d8cab06-6cec-49b1-8fa8-53aa426c38df",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "model_dir: /home/ubuntu/fulldataset/model/GSE135167_mouse_eye_topp1kgenes/best_model/\n",
      "predict_file: /home/ubuntu/fulldataset/GSE135167_mouse_eye.val.h5ad\n",
      "log_dir: /home/ubuntu/scgenai/examples/logs\n",
      "output_dir: /home/ubuntu/scgenai/examples/output\n",
      "seed: 1314521\n",
      "mode: Predict\n",
      "min_cells: 50\n",
      "multiomics: No\n",
      "keepIntermediateFiles: Yes\n",
      "savelog: No\n",
      "savesetting: Yes\n",
      "model_backbone_name: llama\n",
      "model_backbone_size: small\n",
      "context_method: genelist\n",
      "num_bins: 50\n",
      "optimizer: AdamW\n",
      "weight_decay: 0.01\n",
      "depth: 2\n",
      "max_length: 2048\n",
      "batch_size: 1\n",
      "learning_rate: 1e-05\n",
      "num_epochs: 30\n",
      "world_size: 1\n",
      "evaluate_during_training: False\n",
      "hidden_size: 768\n",
      "num_layers: 12\n",
      "num_heads: 12\n",
      "intermediate_size: 1024\n",
      "stride: 1024\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2024-10-22 00:23:36.958 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 3, Total GPUs used: 4\n",
      "2024-10-22 00:23:36.962 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 0, Total GPUs used: 4\n",
      "2024-10-22 00:23:36.970 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 1, Total GPUs used: 4\n",
      "2024-10-22 00:23:36.996 | INFO     | scGenAI.utils.distributed:setup_distributed:26 - Distributed setup complete for GPU 2, Total GPUs used: 4\n",
      "GPU 2 Predicting: 100%|██████████| 379/379 [00:21<00:00, 17.36it/s]\n",
      "GPU 0 Predicting: 100%|██████████| 379/379 [00:22<00:00, 17.17it/s]\n",
      "GPU 3 Predicting: 100%|██████████| 379/379 [00:22<00:00, 17.16it/s]\n",
      "GPU 1 Predicting: 100%|██████████| 379/379 [00:22<00:00, 17.01it/s]\n",
      "2024-10-22 00:24:02.912 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "2024-10-22 00:24:02.912 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "2024-10-22 00:24:02.912 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "2024-10-22 00:24:02.912 | INFO     | scGenAI.utils.distributed:cleanup:33 - Cleaned up DDP environment\n",
      "\u001b[32m2024-10-22 00:24:04.295\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mscGenAI.data.postprocess\u001b[0m:\u001b[36mcombine_prediction_results\u001b[0m:\u001b[36m103\u001b[0m - \u001b[1mCompleted post-processing for prediction\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "pred_config_file = '/home/ubuntu/scgenai/tutorials/yaml/config_prediction_GSE135167_mouse_eye.yaml'\n",
    "run_prediction_from_config(pred_config_file)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7fa3e6fc-55c8-4ef4-b64e-f6347b84fbb4",
   "metadata": {},
   "source": [
    "The prediction output is a CSV file, as defined in the configuration file. It contains the original metadata extracted from the input prediction file (`obs` slot) along with three additional columns: `context_id`, `PredictedFeature`, and `prediction_score`. \n",
    "\n",
    "- `context_id` represents the context used to determine the prediction for the corresponding cell. \n",
    "- `PredictedFeature` is the final predicted feature for the cell using the trained model. \n",
    "- `prediction_score` indicates the confidence level of the prediction, with a maximum value of 1.\n",
    "\n",
    "For example in this tutorials:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "b4c89ab7-232d-41a1-8cb4-f59851b5fe6d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>index</th>\n",
       "      <th>orig.ident</th>\n",
       "      <th>conditions</th>\n",
       "      <th>var.ratio.pca</th>\n",
       "      <th>res.0.8</th>\n",
       "      <th>percent.mito</th>\n",
       "      <th>ClusterNames_0.8</th>\n",
       "      <th>Cell.types</th>\n",
       "      <th>nCount_RNA</th>\n",
       "      <th>nFeature_RNA</th>\n",
       "      <th>ct</th>\n",
       "      <th>cell_barcode</th>\n",
       "      <th>context_id</th>\n",
       "      <th>PredictedFeature</th>\n",
       "      <th>prediction_score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>B6_CGATTGATCTTCAACT</td>\n",
       "      <td>B6</td>\n",
       "      <td>B6</td>\n",
       "      <td>0.973345</td>\n",
       "      <td>3</td>\n",
       "      <td>0.023335</td>\n",
       "      <td>9</td>\n",
       "      <td>Smooth muscle cells</td>\n",
       "      <td>2057</td>\n",
       "      <td>951</td>\n",
       "      <td>Smooth muscle cells</td>\n",
       "      <td>B6_CGATTGATCTTCAACT</td>\n",
       "      <td>3_B6_CGATTGATCTTCAACT_window_0</td>\n",
       "      <td>Smooth muscle cells</td>\n",
       "      <td>0.999994</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>B6_CCGGGATAGCTAAACA</td>\n",
       "      <td>B6</td>\n",
       "      <td>B6</td>\n",
       "      <td>1.007342</td>\n",
       "      <td>7</td>\n",
       "      <td>0.024749</td>\n",
       "      <td>8</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>5374</td>\n",
       "      <td>2296</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>B6_CCGGGATAGCTAAACA</td>\n",
       "      <td>0_B6_CCGGGATAGCTAAACA_window_1</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>0.999985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>B6_AACGTTGCAAGCGAGT</td>\n",
       "      <td>B6</td>\n",
       "      <td>B6</td>\n",
       "      <td>1.003478</td>\n",
       "      <td>3</td>\n",
       "      <td>0.022488</td>\n",
       "      <td>9</td>\n",
       "      <td>Smooth muscle cells</td>\n",
       "      <td>2846</td>\n",
       "      <td>1322</td>\n",
       "      <td>Smooth muscle cells</td>\n",
       "      <td>B6_AACGTTGCAAGCGAGT</td>\n",
       "      <td>1_B6_AACGTTGCAAGCGAGT_window_2</td>\n",
       "      <td>Smooth muscle cells</td>\n",
       "      <td>0.999985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>B6_CCTACACAGCTAGCCC</td>\n",
       "      <td>B6</td>\n",
       "      <td>B6</td>\n",
       "      <td>0.969138</td>\n",
       "      <td>1</td>\n",
       "      <td>0.031937</td>\n",
       "      <td>6</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>4415</td>\n",
       "      <td>1933</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>B6_CCTACACAGCTAGCCC</td>\n",
       "      <td>2_B6_CCTACACAGCTAGCCC_window_3</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>0.999987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>129_B6_GGGCATCCATTCGACA</td>\n",
       "      <td>129</td>\n",
       "      <td>129_B6</td>\n",
       "      <td>0.964243</td>\n",
       "      <td>1</td>\n",
       "      <td>0.033408</td>\n",
       "      <td>6</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>4879</td>\n",
       "      <td>2022</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>129_B6_GGGCATCCATTCGACA</td>\n",
       "      <td>1_129_B6_GGGCATCCATTCGACA_window_4</td>\n",
       "      <td>Stromal cells</td>\n",
       "      <td>0.999987</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                     index orig.ident conditions  var.ratio.pca  res.0.8   \n",
       "0      B6_CGATTGATCTTCAACT         B6         B6       0.973345        3  \\\n",
       "1      B6_CCGGGATAGCTAAACA         B6         B6       1.007342        7   \n",
       "2      B6_AACGTTGCAAGCGAGT         B6         B6       1.003478        3   \n",
       "3      B6_CCTACACAGCTAGCCC         B6         B6       0.969138        1   \n",
       "4  129_B6_GGGCATCCATTCGACA        129     129_B6       0.964243        1   \n",
       "\n",
       "   percent.mito  ClusterNames_0.8           Cell.types  nCount_RNA   \n",
       "0      0.023335                 9  Smooth muscle cells        2057  \\\n",
       "1      0.024749                 8        Stromal cells        5374   \n",
       "2      0.022488                 9  Smooth muscle cells        2846   \n",
       "3      0.031937                 6        Stromal cells        4415   \n",
       "4      0.033408                 6        Stromal cells        4879   \n",
       "\n",
       "   nFeature_RNA                   ct             cell_barcode   \n",
       "0           951  Smooth muscle cells      B6_CGATTGATCTTCAACT  \\\n",
       "1          2296        Stromal cells      B6_CCGGGATAGCTAAACA   \n",
       "2          1322  Smooth muscle cells      B6_AACGTTGCAAGCGAGT   \n",
       "3          1933        Stromal cells      B6_CCTACACAGCTAGCCC   \n",
       "4          2022        Stromal cells  129_B6_GGGCATCCATTCGACA   \n",
       "\n",
       "                           context_id     PredictedFeature  prediction_score  \n",
       "0      3_B6_CGATTGATCTTCAACT_window_0  Smooth muscle cells          0.999994  \n",
       "1      0_B6_CCGGGATAGCTAAACA_window_1        Stromal cells          0.999985  \n",
       "2      1_B6_AACGTTGCAAGCGAGT_window_2  Smooth muscle cells          0.999985  \n",
       "3      2_B6_CCTACACAGCTAGCCC_window_3        Stromal cells          0.999987  \n",
       "4  1_129_B6_GGGCATCCATTCGACA_window_4        Stromal cells          0.999987  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "preconfig = Config(pred_config_file)\n",
    "dfprediction = pd.read_csv(preconfig.outputfile)\n",
    "dfprediction.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0cd6dab7-bb05-4204-bc26-b730ad4461a4",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": false,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": true
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}