transformer.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<torch._C.Generator at 0x10b838ab0>"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from torch.utils.data import DataLoader, TensorDataset\n",
    "import numpy as np\n",
    "import pickle\n",
    "import math\n",
    "from bidict import bidict\n",
    "from tqdm import tqdm\n",
    "from utils.dataset import generate_dataset\n",
    "from utils.Profiler import Profiler\n",
    "from copy import deepcopy\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from SequentialGroup import Column, Group, Node, Link\n",
    "\n",
    "np.random.seed(0)\n",
    "torch.manual_seed(0)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_curve(y, color='#1F77B4', color2=\"#82B0D2\", window_size = 1000, y_bottom=None, y_top=None, label=None):\n",
    "    s = pd.Series(y)\n",
    "    s_ma = s.rolling(window=window_size, min_periods=1).mean()\n",
    "    upper_quartile = s.rolling(window=window_size, min_periods=1).quantile(0.75)\n",
    "    lower_quartile = s.rolling(window=window_size, min_periods=1).quantile(0.25)\n",
    "    # plt.plot(s, c=color2, alpha=0.3, zorder=2)\n",
    "    plt.fill_between(s.index, lower_quartile, upper_quartile, color=color2, zorder=4)\n",
    "    plt.plot(s_ma, color=color, label=label, zorder=6)\n",
    "    plt.ylim(bottom=y_bottom, top=y_top)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "seq1 = ['A', 'B', 'C', 'D', 'E']\n",
    "seq2 = ['X', 'B', 'C', 'D', 'Y']\n",
    "seqs = [seq1, seq2]\n",
    "\n",
    "D_train, _, dictionary = generate_dataset(seqs)\n",
    "data = D_train[:5000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_ = torch.tensor(data)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "embeddings = np.random.uniform(-1, 1, size=(27, 32))\n",
    "mapping = dict({i: embeddings[i] for i in range(len(embeddings))})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## LSTM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class SeL_LSTM(nn.Module):\n",
    "    def __init__(self, embeddings: torch.Tensor, num_layers, hidden_size, seq_length, dropout=0.1):\n",
    "        super(SeL_LSTM, self).__init__()\n",
    "        self.seq_length = seq_length\n",
    "        \n",
    "        embed = torch.FloatTensor(embeddings)\n",
    "        self.embedding = nn.Embedding(num_embeddings=embed.shape[0], embedding_dim=embed.shape[1], _weight=embed)\n",
    "        \n",
    "        self.lstm = nn.LSTM(input_size=embed.shape[1], hidden_size=hidden_size, num_layers=num_layers, dropout=dropout, batch_first=True)\n",
    "        self.decoder = nn.Linear(hidden_size, embed.shape[0])\n",
    "\n",
    "    def forward(self, src):\n",
    "        src = self.embedding(src)\n",
    "        lstm_out, _ = self.lstm(src)\n",
    "        output = self.decoder(lstm_out[:, -1, :])\n",
    "        return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "num_layers = 2\n",
    "num_heads = 4\n",
    "seq_length = 4\n",
    "dropout = 0.1\n",
    "\n",
    "# Initialize model, loss function, and optimizer\n",
    "model = SeL_LSTM(embeddings, num_layers, num_heads, seq_length, dropout)\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  6%|▌         | 281/4996 [00:00<00:04, 955.47it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.1190199851989746\n",
      "3.341696262359619\n",
      "2.999910831451416\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 12%|█▏        | 617/4996 [00:00<00:04, 1078.02it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.214792013168335\n",
      "2.4286561012268066\n",
      "3.273164749145508\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 17%|█▋        | 838/4996 [00:00<00:03, 1074.98it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.6150550842285156\n",
      "3.0552492141723633\n",
      "2.543056011199951\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 23%|██▎       | 1154/4996 [00:01<00:03, 1018.88it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.0703999996185303\n",
      "2.5271387100219727\n",
      "2.8212177753448486\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 29%|██▉       | 1458/4996 [00:01<00:03, 934.10it/s] "
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.254535436630249\n",
      "2.893503189086914\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 33%|███▎      | 1661/4996 [00:01<00:03, 969.66it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.5328187942504883\n",
      "2.560659885406494\n",
      "2.2366414070129395\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 40%|███▉      | 1975/4996 [00:01<00:02, 1018.63it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.4862120151519775\n",
      "2.1990742683410645\n",
      "2.486154317855835\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 46%|████▌     | 2303/4996 [00:02<00:02, 1066.54it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.200063705444336\n",
      "2.4062721729278564\n",
      "2.0461935997009277\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 51%|█████     | 2523/4996 [00:02<00:02, 1079.52it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.619107246398926\n",
      "2.031132221221924\n",
      "2.1533584594726562\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 57%|█████▋    | 2861/4996 [00:02<00:01, 1109.27it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9692069292068481\n",
      "2.6818671226501465\n",
      "1.7724192142486572\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 64%|██████▍   | 3202/4996 [00:03<00:01, 1125.56it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.3915317058563232\n",
      "1.6749298572540283\n",
      "2.331885814666748\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 69%|██████▊   | 3429/4996 [00:03<00:01, 1129.90it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.6520954370498657\n",
      "1.9644815921783447\n",
      "1.5084941387176514\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 76%|███████▌  | 3772/4996 [00:03<00:01, 1129.68it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9149949550628662\n",
      "1.4567431211471558\n",
      "1.8789716958999634\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 82%|████████▏ | 4117/4996 [00:03<00:00, 1138.40it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.4228394031524658\n",
      "1.8180888891220093\n",
      "1.3752927780151367\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 87%|████████▋ | 4345/4996 [00:04<00:00, 1128.91it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.728172779083252\n",
      "1.3883848190307617\n",
      "1.6054695844650269\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 94%|█████████▍| 4689/4996 [00:04<00:00, 1139.54it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.2565319538116455\n",
      "1.6542850732803345\n",
      "1.1492356061935425\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4996/4996 [00:04<00:00, 1073.30it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.8402352333068848\n",
      "1.1433801651000977\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "losses = []\n",
    "accuracies1 = []\n",
    "profiler = Profiler(100)\n",
    "n_span = seq_length\n",
    "topk = 1\n",
    "for i in tqdm(range(n_span, len(data_))):\n",
    "    src, tgt = data_[i-n_span:i], data_[i:i+1]\n",
    "\n",
    "    optimizer.zero_grad()\n",
    "    pred: torch.Tensor = model(src.unsqueeze(0))\n",
    "    loss = criterion(pred.view(-1, pred.shape[-1]), tgt.view(-1))\n",
    "    # loss = criterion(output, tgt.unsqueeze(0))\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "\n",
    "\n",
    "    losses.append(float(loss))\n",
    "    if i%100 == 0:\n",
    "        print(losses[-1])\n",
    "\n",
    "    # labels = torch.argmax(pred.detach(), dim=-1)\n",
    "    _, labels = torch.topk(pred.detach(), topk)\n",
    "    anticipations = set(labels.squeeze(0).tolist())\n",
    "    acc = profiler.observe(anticipations, int(tgt))\n",
    "    \n",
    "    accuracies1.append(float(acc))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'event')"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 200x160 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(2,1.6))\n",
    "plt.subplots_adjust(left=0.3, right=1.0, top=0.95, bottom=0.3)\n",
    "plot_curve(accuracies1[:50000], y_bottom=0, y_top=1, label=\"all\")\n",
    "plt.axhline(0.5, color='#999999', alpha=0.5, linestyle='--', zorder=5)\n",
    "# plot_curve(accuracies1_2[:50000], color=\"#2CA02C\", color2=\"#8ECFC9\", y_bottom=0, y_top=1, label=\"Top1\")\n",
    "# plt.legend(loc='upper right')\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.xlabel(\"event\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Transformer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "class SeL_Transformer(nn.Module):\n",
    "    def __init__(self, embeddings: torch.Tensor, num_layers, num_heads, seq_length, dropout=0.1):\n",
    "        super(SeL_Transformer, self).__init__()\n",
    "        self.seq_length = seq_length\n",
    "        \n",
    "        hidden_size = embeddings.shape[1]\n",
    "        \n",
    "        embed = torch.FloatTensor(embeddings)\n",
    "        self.embedding = nn.Embedding(num_embeddings=embed.shape[0], embedding_dim=embed.shape[1], _weight=embed)\n",
    "        \n",
    "        self.positional_encoding = nn.Parameter(torch.zeros(1, seq_length, hidden_size))\n",
    "        encoder_layer = nn.TransformerEncoderLayer(d_model=hidden_size, nhead=num_heads, dropout=dropout)\n",
    "        self.transformer_encoder = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)\n",
    "        self.decoder = nn.Linear(hidden_size, embed.shape[0])\n",
    "\n",
    "    def forward(self, src):\n",
    "        src = self.embedding(src) + self.positional_encoding[:, :self.seq_length, :]\n",
    "        output = self.transformer_encoder(src)\n",
    "        output = self.decoder(output[:, -1, :])\n",
    "        return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/homebrew/Caskroom/miniforge/base/envs/PyNARS/lib/python3.9/site-packages/torch/nn/modules/transformer.py:306: UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.self_attn.batch_first was not True(use batch_first for better inference performance)\n",
      "  warnings.warn(f\"enable_nested_tensor is True, but self.use_nested_tensor is False because {why_not_sparsity_fast_path}\")\n"
     ]
    }
   ],
   "source": [
    "num_layers = 2\n",
    "num_heads = 4\n",
    "seq_length = len(seq1)\n",
    "dropout = 0.1\n",
    "\n",
    "# Initialize model, loss function, and optimizer\n",
    "model = SeL_Transformer(embeddings, num_layers, num_heads, seq_length, dropout)\n",
    "criterion = nn.CrossEntropyLoss()\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  3%|▎         | 129/4995 [00:00<00:30, 159.14it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.8090460300445557\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  4%|▍         | 219/4995 [00:01<00:27, 171.87it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4.978393077850342\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  7%|▋         | 330/4995 [00:02<00:27, 172.18it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6236518621444702\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  8%|▊         | 418/4995 [00:02<00:27, 166.03it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.7606031894683838\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 10%|█         | 521/4995 [00:03<00:27, 159.99it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.050177574157715\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 13%|█▎        | 629/4995 [00:03<00:25, 171.76it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.5179897546768188\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 14%|█▍        | 716/4995 [00:04<00:26, 159.79it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.614595353603363\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 16%|█▋        | 822/4995 [00:05<00:24, 169.69it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4.55616569519043\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 19%|█▊        | 928/4995 [00:05<00:23, 173.83it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8705958724021912\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 20%|██        | 1021/4995 [00:06<00:22, 174.48it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9572181701660156\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 23%|██▎       | 1128/4995 [00:06<00:23, 161.16it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.970343828201294\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 24%|██▍       | 1215/4995 [00:07<00:23, 160.22it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.189523935317993\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 26%|██▋       | 1316/4995 [00:08<00:27, 134.41it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7848700284957886\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 28%|██▊       | 1417/4995 [00:08<00:22, 155.86it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.545750617980957\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 30%|███       | 1523/4995 [00:09<00:21, 161.98it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.4541842937469482\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 32%|███▏      | 1613/4995 [00:09<00:19, 171.18it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.7931586503982544\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 35%|███▍      | 1724/4995 [00:10<00:18, 178.73it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8842716217041016\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 36%|███▋      | 1816/4995 [00:11<00:19, 160.82it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9778928756713867\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 39%|███▊      | 1927/4995 [00:11<00:18, 167.82it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0127936601638794\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 40%|████      | 2016/4995 [00:12<00:18, 164.35it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.203900098800659\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 42%|████▏     | 2116/4995 [00:13<00:19, 149.08it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9706125259399414\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 45%|████▍     | 2223/4995 [00:13<00:16, 171.79it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5.392150402069092\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 46%|████▋     | 2314/4995 [00:14<00:18, 141.33it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.1614911556243896\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 48%|████▊     | 2414/4995 [00:14<00:17, 146.50it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.850930690765381\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 51%|█████     | 2524/4995 [00:15<00:16, 150.76it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.614226222038269\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 52%|█████▏    | 2611/4995 [00:16<00:15, 152.43it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.6079767942428589\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 55%|█████▍    | 2723/4995 [00:16<00:15, 150.30it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8378205895423889\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 57%|█████▋    | 2826/4995 [00:17<00:13, 164.03it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4.498952388763428\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 58%|█████▊    | 2921/4995 [00:18<00:14, 143.36it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.334376335144043\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 60%|██████    | 3020/4995 [00:18<00:12, 158.69it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9621760845184326\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 63%|██████▎   | 3124/4995 [00:19<00:11, 162.32it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0163862705230713\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 64%|██████▍   | 3215/4995 [00:20<00:10, 169.10it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5.109671592712402\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 66%|██████▋   | 3321/4995 [00:20<00:10, 164.99it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.6314192414283752\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 68%|██████▊   | 3409/4995 [00:21<00:10, 154.87it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.8865147829055786\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 71%|███████   | 3524/4995 [00:22<00:09, 150.75it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8670600652694702\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 72%|███████▏  | 3620/4995 [00:22<00:09, 144.63it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.9976520538330078\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 75%|███████▍  | 3724/4995 [00:23<00:07, 167.99it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5384445190429688\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 76%|███████▋  | 3812/4995 [00:23<00:07, 165.19it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.8977537155151367\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 78%|███████▊  | 3918/4995 [00:24<00:06, 161.77it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9206941723823547\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 81%|████████  | 4023/4995 [00:25<00:05, 167.61it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.7029787302017212\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 82%|████████▏ | 4111/4995 [00:25<00:05, 169.78it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8605079054832458\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 84%|████████▍ | 4215/4995 [00:26<00:04, 160.78it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.6895630359649658\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 86%|████████▋ | 4320/4995 [00:26<00:04, 163.74it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.2643780708312988\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 89%|████████▊ | 4427/4995 [00:27<00:03, 168.60it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.884316325187683\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 90%|█████████ | 4515/4995 [00:28<00:02, 168.78it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.2683825492858887\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 93%|█████████▎| 4621/4995 [00:28<00:02, 171.57it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.7402046918869019\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 94%|█████████▍| 4714/4995 [00:29<00:01, 177.88it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.676694929599762\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 97%|█████████▋| 4825/4995 [00:29<00:00, 177.12it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4.979320049285889\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 98%|█████████▊| 4914/4995 [00:30<00:00, 164.94it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9476185441017151\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4995/4995 [00:30<00:00, 161.48it/s]\n"
     ]
    }
   ],
   "source": [
    "losses = []\n",
    "accuracies2 = []\n",
    "profiler = Profiler(100)\n",
    "n_span = seq_length\n",
    "topk = 1\n",
    "for i in tqdm(range(n_span, len(data_))):\n",
    "    src, tgt = data_[i-n_span:i], data_[i:i+1]\n",
    "\n",
    "    optimizer.zero_grad()\n",
    "    pred: torch.Tensor = model(src.unsqueeze(0))\n",
    "    loss = criterion(pred.view(-1, pred.shape[-1]), tgt.view(-1))\n",
    "    # loss = criterion(output, tgt.unsqueeze(0))\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "\n",
    "\n",
    "    losses.append(float(loss))\n",
    "    if i%100 == 0:\n",
    "        print(losses[-1])\n",
    "\n",
    "    # labels = torch.argmax(pred.detach(), dim=-1)\n",
    "    _, labels = torch.topk(pred.detach(), topk)\n",
    "    anticipations = set(labels.squeeze(0).tolist())\n",
    "    acc = profiler.observe(anticipations, int(tgt))\n",
    "    \n",
    "    accuracies2.append(float(acc))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'event')"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 200x160 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(2,1.6))\n",
    "plt.subplots_adjust(left=0.3, right=1.0, top=0.95, bottom=0.3)\n",
    "plot_curve(accuracies2[:50000], y_bottom=0, y_top=1, label=\"all\")\n",
    "plt.axhline(0.5, color='#999999', alpha=0.5, linestyle='--', zorder=5)\n",
    "# plot_curve(accuracies1_2[:50000], color=\"#2CA02C\", color2=\"#8ECFC9\", y_bottom=0, y_top=1, label=\"Top1\")\n",
    "# plt.legend(loc='upper right')\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.xlabel(\"event\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SeL-NAL"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_model(n_columns, n_nodes):\n",
    "    # Create the model\n",
    "    g = Group(n_columns, n_nodes)\n",
    "    g.thresh = 0.8\n",
    "    g.p_plus = 2 #1.95\n",
    "    g.p_minus = 0.99 #0.9\n",
    "    g.p_minus2 = 0.01\n",
    "\n",
    "    for idx, column in enumerate(g.columns):\n",
    "        column.mark = idx\n",
    "    return g\n",
    "\n",
    "def test_model(g: Group, data, top1=False):\n",
    "    print(len(data))\n",
    "    print(data[:20])\n",
    "    accuracies = []\n",
    "    n_anticipations = []\n",
    "\n",
    "    profiler = Profiler(100)\n",
    "    for i, (idx, idx_next) in enumerate((tqdm(zip(data[:-1], data[1:]), total=len(data)-1))):\n",
    "        candidates: set[Node] = g.activate(idx)\n",
    "        anticipations = set()\n",
    "        if not top1:\n",
    "            for candidate in candidates:\n",
    "                anticipations.add(candidate.column.mark)\n",
    "        else:\n",
    "            if len(candidates) > 0:\n",
    "                candidate = max(candidates, key=lambda node: node.activity_pred)\n",
    "                anticipations.add(candidate.column.mark)\n",
    "\n",
    "        ''' compute the accuracy '''\n",
    "        acc = profiler.observe(anticipations, idx_next)\n",
    "        accuracies.append(acc)\n",
    "        ''' count the number of valid anticipations '''\n",
    "        a = np.array(g.columns[idx].get_activities(), dtype=bool)\n",
    "        if not all(a):\n",
    "            n_anticipations.append((len(anticipations), sum(a)))\n",
    "        # n_anticipations.append(len(anticipations))\n",
    "        \n",
    "    return accuracies, g, n_anticipations"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5000\n",
      "[3, 5, 2, 6, 0, 6, 13, 19, 3, 5, 2, 6, 0, 11, 14, 1, 4, 5, 2, 6]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4999/4999 [00:00<00:00, 10466.04it/s]\n"
     ]
    }
   ],
   "source": [
    "g1 = create_model(26, 10)\n",
    "accuracies3, _, _ = test_model(g1, data, top1=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'event')"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAANMAAACsCAYAAADhc5nNAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjkuMCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy80BEi2AAAACXBIWXMAAA9hAAAPYQGoP6dpAAAf20lEQVR4nO3deVQUV/o38G9V9Q50swqCLBo17gu4YYwmhgSXJEeTnxOdTHR81XkTNS4YjeQk42Q5IWOcmWSME5M/1DM5kxjjb5IzEx3zEhScUcaoEYIKJq4QpUFAummg17rvHw0tTTdbU9A0Pp9zONq3bhVPQT3cqlt1b3GMMQZCSLfx/g6AkP6CkokQiVAyESIRSiZCJELJRIhEKJkIkQglEyESoWQiRCKUTIRIhJKJEIn4NZmOHz+OJ554ArGxseA4Dl999VWH6+Tm5iI5ORlKpRJDhw7Fvn37ejxOQjrDr8lUX1+P8ePHY9euXZ2qf+3aNcyfPx8PP/wwCgoKsGHDBqxcuRLffPNND0dKSMe4vvKgK8dx+PLLL7FgwYI267z88ss4dOgQzp8/7ypbvHgxamtrceTIkV6IkpC2yfwdQFfk5+cjLS3NrSw9PR0bNmxocx2LxQKLxeL6LIoiampqEBERAY7jeipU0k8wxlBXV4fY2FjwfPsncgGVTHq9HtHR0W5l0dHRMBqNaGxshFqt9lgnKysLr7/+em+FSPqpsrIyDBo0qN06AZVMvsjMzERGRobrs8FgQEJCAsrKyqDVav0YGQkERqMR8fHxCAkJ6bBuQCVTTEwMKioq3MoqKiqg1Wq9tkoAoFQqoVQqPcq1Wi0lE+m0zlwSBNR9ptTUVOTk5LiVZWdnIzU11U8REXKXX5PJZDKhoKAABQUFAJxd3wUFBSgtLQXgPEVbunSpq/7zzz+Pq1evYsuWLSgpKcFf/vIXHDhwABs3bvRH+IS4Y3507NgxBsDja9myZYwxxpYtW8ZmzZrlsc6ECROYQqFgQ4YMYXv37u3S9zQYDAwAMxgM0uwE6de6crz0mftMvcVoNEKn08FgMNA1E+lQV46XgLpmIqQvo2QiRCKUTIRIhJKJEIlQMhEiEUomQiRCyUSIRCiZCJEIJRMhEqFkIkQilEyESISSiRCJUDIRIhFKJkIkQslEiEQomQiRiE/JdOzYManjICTg+ZRMc+bMwX333Ye33noLZWVlUsdESEDyKZlu3ryJtWvX4uDBgxgyZAjS09Nx4MABWK1WqeMjJGD4lEyRkZHYuHEjCgoKcOrUKQwfPhyrV69GbGws1q1bh8LCQqnjJKTP63YHRHJyMjIzM7F27VqYTCbs2bMHKSkpePDBB3HhwgUpYiQkIPicTDabDQcPHsS8efOQmJiIb775Bh988AEqKipw+fJlJCYmYtGiRVLGSkif5tNUXy+++CI+++wzMMbw3HPPYeXKlRgzZoxbHb1ej9jYWIiiKFmwUqCpvkhXdOV48Wmu8YsXL2Lnzp146qmnvM7jDTivq6gLndxLaBJKQtrR45NQZmVlYc+ePR7le/bswe9//3tfNklIwPMpmT766COMGDHCo3z06NHYvXt3t4MiJBD5lEx6vR4DBw70KI+KikJ5eXm3gyIkEPmUTPHx8Thx4oRH+YkTJxAbG9vtoAgJRD715q1atQobNmyAzWbD7NmzAQA5OTnYsmULNm3aJGmAhAQKn5Jp8+bNqK6uxurVq13P46lUKrz88svIzMyUNEBCAkW3usZNJhOKi4uhVqsxbNiwNu859SXUNU66osdv2jYLDg7G5MmTu7MJQvoNn5PpzJkzOHDgAEpLSz2GXvz973/vdmCEBBqfevP279+P6dOno7i4GF9++SVsNhsuXLiAo0ePQqfTSR0jIQHBp2R6++238ac//Qn//Oc/oVAo8P7776OkpAS/+MUvkJCQIHWMhAQEn5LpypUrmD9/PgBAoVCgvr4eHMdh48aN+PjjjyUNkJBA4dM1U1hYGOrq6gAAcXFxOH/+PMaOHYva2lo0NDRIGiDxv9tGM/71wy04RAZw3dwYA4bFhGDG8AGSxNaX+JRMM2fORHZ2NsaOHYtFixZh/fr1OHr0KLKzs/HII49IHSPxE7tDxD/P/YzzNw0+rW+xizCabXCIDBq5gBCVDBzH4ec7DSitrgcAJCeGY0Rs/7jO9uk+U01NDcxms2vw3/bt23Hy5EkMGzYMr776KsLCwnoiVknQfabOOXutGt/fqIHeYHYrZ4zBbBdhtYuwORhExmAXGSxNZRa7iHqrHeUGC+402tzWVct5xIeqMVCnQrhGDq1aDhnPYcRA5+/hvgEhSE4KbzOm765W4UaVMwkVMh7pY2OhkgsS77m7rhwvXU4mu92OTz/9FOnp6YiOju5WoP5AydS261UmFNy4AwAo+rkWAGCxOXDTYEZFnQW3TVbUNthgEzt/yGjkAmQCh3qLAw4vh5pGLiAiWI6EMDWGRAQhpZ1kao6pmUzgMHJg51s1kTEIPIe2jniNUkDa6IHgubvnsj2aTACg0WhQXFyMxMTErq7q1a5du/Duu+9Cr9dj/Pjx2LlzJ6ZMmeK17r59+7B8+XK3MqVSCbPZ7LV+a/dSMt2804CC0judrv/d1WoYGm0wNNpQ22jDLYMFlXUWeDtA5AIHucCD5wAZz0Mp46EQOChkPFRyAQOCFYjVqVwth0Nk0BstKKttRJXJijsNVlgd7luW8xwSw9VQKwTwHAeFwEGjkMEhMtSZ7TBZ7Kiz2sFzHEKUMkQEyaFTyRGkEMDzzgRgjEFkAM9x4DmA553/chwHh8hgNNtR22CDyWKHUsbDLjJUmqyorrdC4DmceHk2woIUrph6/AmIKVOmoKCgQJJk+vzzz5GRkYHdu3dj6tSpeO+995Ceno5Lly5hwADvF6larRaXLl1yfea47l4V911lNfUouWXssJ4oMtRZnAdcvdWBeosdRT/XNl23OMtNFjtEBoSoZNCpZAhWyiBwQJ3FgUqTBVUmK7w1OqFqOWJ1SgwIUSJco4BWJYPAd+1nLvAc4kJViAtVAXAe9M2xlRvN+Ol2PQyNdlyu6lwHlh4W/HS7SyF0SCnju7xfLfmUTKtXr0ZGRgbKysqQkpKCoKAgt+Xjxo3r9Lb++Mc/YtWqVa7WZvfu3Th06BD27NmDrVu3el2H4zjExMT4EjoAQBAEOBwOOBwOr9vm+bt3DLzV8Va3wtCICz+33wpwPI9gpQyTBkfg++vVMDS0PWkn17Td//x4G2AivP2O660OlFTU43pNI2obbV4TwRuj2Y6bbSxTCBxC1XKEqmWICFYiPkyNEKUMznOj5m/A4HGuxLW4y9JUVy5wCFbKXcV3XPvLgeM4qOQCVDIeA4LlGDcwGLeMFlQYLbCJDIyh6frLAYF3tkQhSgHBSgEiAwxmO+402GA029FgdYDBmaA8x4HjAJE5T+tE0dlSMTg7IkNUMoSqZQhRylwtY1SQAlHBcgg813RcdO7335pPybR48WIAwLp161xlHMeBMeZsTjsZgNVqxdmzZ92eNOd5HmlpacjPz29zPZPJhMTERIiiiOTkZLz99tsYPXq017oWiwUWi8X12Wg0Yu7cubh48aLXB3N1Oh2GDRvm+lxYWAhRdF5YVxrNaHlWHB8dgeRxo3H2ejW+KSqHpv4meOZ9NiZRUKBB7bzG1BvM+LH4Ijhm99rTLPJyNGiicbvOipsGMxrr62C2i7A4ALsIOJjz31ovuagUALXg/FfV9K9OAWgVQIiCh1UdCaPZDqPZDnOjCUwUESQDotRAbBCglTNwnA2AHfXBka7tqsxVEBzup9IK4e5BZwu9e5Yi1FdCKZoxITYcAn83SKPahuJbBli18a7kExqqwFudnQrBMhHDW10y1QfFApzzVFFhuQO5ramVVgNo1c/VoBkIxsua6tZCbqtDc19+8++N4xwAHGjQhIHxzkSXWw1QWG8DDLh4/gcoZXf3q+Wx0xGfkunatWu+rOahqqoKDofDoyMjOjoaJSUlXte5//77sWfPHowbNw4GgwE7duzA9OnTceHCBQwaNMijflZWFl5//XW3sscff7zLsZbcMsDQaHMd/AzAtVoHyuxaXGjRddxW49CcgxwHFJbegZoDGm0MPxmAKjNgsgFWEXCIgJ3ZUGf72eOawpuBGg7DY8MRrVUiSCEgqLESvHj3AJa1aNIYx8MSonF95uqsEBydP1iaCRyg1SgwdlCoq2zSpJGu/1+5osCdO56ttFYtx9T7IjFx4ggIgjNBrl+/jqqqKgDArTuNuHbbhJZn7UqBB5paaZmdg8zh+efH3kaTLHNrzt3XUwocWNMfA5mDh8ABCpkAeTdO8/w6O9GtW7cQFxeHkydPIjU11VW+ZcsW5OXl4dSpUx1uw2azYeTIkViyZAnefPNNj+XeWqakpCTcvn3b6wWlt9M8u0PE9sMXvQfA8eDQlERttEot6zarMDbi2x+rYba1vY5C4BAXqkaYWgaVnIdScJ7TN39pVTIEKQSAa3Gu3xSDKDI8PCoG0+6LdNtm80EMAP8u0SOvpMJ18d5evKxp/sPlDw5BtE7tUbXldkVRRHuHFc/zrutcKer+7+lSXNIbAXAAxyEiWIH/+9DQNrfZlRiMRiPCw8N7rgPir3/9a7vLly5d2qntREZGQhAEVFRUuJVXVFR0+ppILpdj4sSJuHz5stflSqXS43TO4XBAEAS3A6AtgiDgRnWD+zVBK65fQzt1WjI02vD/SqpgdTDoVDIMjtAgSClzXQALHAeljEd4kNytm7Yt0ToVfvPQsA7rtfbgiBg8OML3a8+2tPxj1Bt1fzFtcKe30dUYOnOMNPMpmdavX+/22WazoaGhAQqFAhqNptPJpFAokJKSgpycHCxYsACA869ETk4O1q5d26ltOBwOFBUVYd68eV3ah65ofeOyOxhjOHmtBlYHw4AQBeaOHACZ4NMjkhB4DpmPj+7XvZmBxKdk8nY+/NNPP+GFF17A5s2bu7StjIwMLFu2DJMmTcKUKVPw3nvvob6+3tW7t3TpUsTFxSErKwsA8MYbb2DatGkYOnQoamtr8e677+LGjRtYuXKlL7vSKa1vFnZH6Z1G3DJYIHDAQ0Mj202kqBAlVsxq+3RF4DhKpD6kWyNtWxo2bBjeeecd/OpXv2qz88CbZ555Brdv38Zvf/tb6PV6TJgwAUeOHHF1SpSWlro1w3fu3MGqVaug1+sRFhaGlJQUnDx5EqNGjZJqV9yYbQ5UGjvfMjlEhpsGM0xmO+yi83GbRpsDNfU2GMw2NDZdI42J1SJE5fzxL31gCKJCPHsWlXIeQhdOg4h/SdoBUVBQgJkzZ8Jo7Pgmo7909QmI/Mu38e0FfYf1RMZQrDfhh5tGNNjavzUQHaLEYyOj8KvUwRgU7rxeIn1Tjz8B8Y9//MPtM2MM5eXl+OCDD/DAAw/4ssk+61J5x38Y6i125F6uht7o7DXUyAUkhKvhEBk4znk/JlwjR5hGgfWPDYdWJYdc4KHTKDrYMgkkPiVTc2dBM47jEBUVhdmzZ+MPf/iDFHH1GcZWTz63VmG0IPvSbVjsIuQ8h19Ojcf/eWAIBoaq0Wizo8Fyt5UKVsoQopa3szUSyHxKpr72zqWeZGgnmX6qNOHE1Ro4GBARJMeORePx0P0DXJ0CCpkCXm7JkH6KTtbbYbN7/6PRaHUg//odXKt2PpSZGKbG4smD8PCIwBuSQqTjU1fR008/7fXVMdu3b+9Xr9701ipduV2P/y0sx7XqBnAAkuN1eOT+SMwZS3Os3+t8Sqbjx497vUk6d+5cHD9+vNtB9RWGRvcnSYtuGZF7uRoWu4hwjRxPjovBxEE6cByHpMigNrZC7hU+neaZTCYoFJ49UXK5vE93i3dVjenuM321jTacKa0FAIyP0yJ5kM71TFtChIZunhLfWqaxY8fi888/9yjfv39/j9089YeWDyMX/GyAyIBBoSqkxOvcHg6dMiTSy9rkXuNTy/Taa6/hqaeewpUrV9xeKfPZZ5/hiy++kDRAf7pdZwYHoM5ix9WmzobkeJ1HKxQfrvGyNrnX+JRMTzzxBL766iu8/fbbOHjwINRqNcaNG4dvv/0Ws2bNkjpGv6mpt4IBKKkwgTFgoFaJqGD3x354DghW0b0j0o2u8fnz57tmde2v5Lxz9HBzF/j90cEedcbF991pzUjv8uma6fTp014H7p06dQpnzpzpdlB9hc3BXPMMCByHhDDPO7DNc74R4lMyrVmzBmVlZR7lN2/exJo1a7odVF9xs7YB5U3P2w3UKSH3MlyCTvFIM5+S6eLFi0hOTvYonzhxIi5ebGN4dwBiDNA3Db+I0Xp/K6K3oRPk3uRTMimVSo+h5gBQXl4Omaz/PKEkiqLrSfCYEJXbMp4DUpLCfR4lS/ofn46Exx57DJmZmTAY7s7KU1tbi1deeQWPPvqoZMH5E2MMdxrtMNtFCByHyGDPm9TBqv7zh4N0n09Hw44dOzBz5kwkJiZi4sSJAJwDA6Ojo/HJJ59IGqC/WOx3W6UBIQqPmT5FBrcJFgnxKZni4uLwww8/4G9/+xsKCwuhVquxfPlyLFmyBHJ5/zjAqk2WDq+XBtHNWtKCz+cpQUFBmDFjBhISElwviP7Xv/4FAHjyySelic6Pquosd6+XtCqP5UOigjHASzm5d/mUTFevXsXChQtRVFTkNi1ys67Mz9xXnb5WjXqrcz8igzyvlx4eSWOXiDufOiDWr1+PwYMHo7KyEhqNBufPn0deXh4mTZqE3NxciUPsfXaHiO+b3lMUopRB0WLu6dhQNdY9dj9iw+gUj7jzqWXKz8/H0aNHERkZCZ7nIQgCZsyYgaysLKxbtw7nzp2TOs5eZbGLqG56Y0O4xv0acPaoGOjUNBEK8eRTy+RwOBASEgLAOcXxrVu3AACJiYlu700KVA1WO2rqnaNsw4PuJhMHIDKYbtIS73xqmcaMGYPCwkIMHjwYU6dOxfbt26FQKPDxxx9jyJAhUsfY62obrKhtGrIe1jQdF8cBiRFBNLsQaZNPyfTqq6+ivt75Tp033ngDjz/+OB588EFERER4HTQYaMxWh2v+B53a+SPiACTS0HTSDp+SKT093fX/oUOHoqSkBDU1NQgLC+sXw7erTBZYHQwcAG3Tg6zO11dSq0TaJtnzMOHhbb8lO9D8WFEHwDlpZMsXZg2O8hzPREgzerjMi6u3nYMBW57ixYapEUrTGZN20CPPXpQbGgE43zIOAOCAQXRfiXSAksmLW00vN9M1JRNjwOABdIpH2kfJ5EVlnfOZvObTPLVcwJCoEH+GRAIAJVMr9RbnnA+A8zSP44CJSeEeQzAIaY2SqZUfypwDHhUCD5WMB2PA+PhQ/wZFAgIlUyuX9M5ucZ1aBo7jEKNTITKEhlqQjlEytfJjpTOZQtVycABGxYX6NR4SOCiZWimtbr7HJAcDkECjaUknUTK1crO2+R6TsyePTvFIZ1EytcAYc7vHpJLzUMnpR0Q6h46UFvRGMyx2ERwH6FQyjI/vHw/ukt5BydTClUrnsBKt0tmTNzSa5hEnnUfJ1MLVKhMA5ymeWi4gMZI6H0jn9Ylk2rVrF5KSkqBSqTB16lR899137db/4osvMGLECKhUKowdOxaHDx+WJI4rlc3JJMPoOB0Evk/8eEiA8PvR8vnnnyMjIwPbtm3D999/j/HjxyM9PR2VlZVe6588eRJLlizBihUrcO7cOSxYsAALFizA+fPnux3L1SrnaV6oWg65zO8/GhJgOMYY67haz5k6dSomT56MDz74AIBzsvz4+Hi8+OKL2Lp1q0f9Z555BvX19fj6669dZdOmTcOECROwe/fuDr+f0WiETqeDwWCAVnv3mqispgEPbj8GAJg3agAWTY7HwyNjurt7JMC1dbx449fBgVarFWfPnkVmZqarjOd5pKWlIT8/3+s6+fn5yMjIcCtLT0/HV1995bW+xWKBxXL3renNLxto/Vb47KJy1/+DlTI4bNZ+9eZ44pvmY6AzbY5fk6mqqgoOhwPR0e6zo0ZHR6OkpMTrOnq93mt9vV7vtX5WVhZef/11j/L4+HiPsv/J+D3GzZwLAFi7cikunc7r1H6Q/q+urg46na7dOv1+2HpmZqZbSyaKImpqahAREeFxD8loNCI+Ph5lZWVY/+0/ejtUSbXcl45OT/o6f+4LYwx1dXWIjY3tsK5fkykyMhKCIHi8OK2iogIxMd6vV2JiYrpUX6lUQql0nzgyNDS03bi0Wm3AH4DNaF+6r6MWqZlfu6wUCgVSUlKQk5PjKhNFETk5OUhNTfW6Tmpqqlt9AMjOzm6zPiG9hvnZ/v37mVKpZPv27WMXL15kv/nNb1hoaCjT6/WMMcaee+45tnXrVlf9EydOMJlMxnbs2MGKi4vZtm3bmFwuZ0VFRd2OxWAwMADMYDB0e1v+RvvS+/yeTIwxtnPnTpaQkMAUCgWbMmUK++9//+taNmvWLLZs2TK3+gcOHGDDhw9nCoWCjR49mh06dEiSOMxmM9u2bRszm82SbM+faF96n9/vMxHSX9BtfkIkQslEiEQomQiRCCUTIRKhZGrS1WEgPS0rKwuTJ09GSEgIBgwYgAULFni8ldFsNmPNmjWIiIhAcHAwnn76aY8b2qWlpZg/fz40Gg0GDBiAzZs3w263u9XJzc1FcnIylEolhg4din379vXovr3zzjvgOA4bNmwI+H1x4+/uxL5g//79TKFQsD179rALFy6wVatWsdDQUFZRUeG3mNLT09nevXvZ+fPnWUFBAZs3bx5LSEhgJpPJVef5559n8fHxLCcnh505c4ZNmzaNTZ8+3bXcbrezMWPGsLS0NHbu3Dl2+PBhFhkZyTIzM111rl69yjQaDcvIyGAXL15kO3fuZIIgsCNHjvTIfn333XcsKSmJjRs3jq1fvz6g96U1SibG2JQpU9iaNWtcnx0OB4uNjWVZWVl+jMpdZWUlA8Dy8vIYY4zV1tYyuVzOvvjiC1ed4uJiBoDl5+czxhg7fPgw43nedQOcMcY+/PBDptVqmcViYYwxtmXLFjZ69Gi37/XMM8+w9PR0yfehrq6ODRs2jGVnZ7NZs2a5kikQ98Wbe/40r3kYSFpamquso2Eg/tA8dKT5pXJnz56FzWZzi3vEiBFISEhwxZ2fn4+xY8e6PWWfnp4Oo9GICxcuuOq03EZznZ7Y9zVr1mD+/Pke3y8Q98Wbfv/UeEd8GQbS20RRxIYNG/DAAw9gzJgxAJxDURQKhcdDuy2Ho7Q1XKV5WXt1jEYjGhsboVarJdmH/fv34/vvv8fp06c9lgXavrTlnk+mQLBmzRqcP38e//nPf/wdik/Kysqwfv16ZGdnQ6Xqv5N63vOneb4MA+lNa9euxddff41jx45h0KBBrvKYmBhYrVbU1ta61W8Zd1vDVZqXtVdHq9VK9pf87NmzqKysRHJyMmQyGWQyGfLy8vDnP/8ZMpkM0dHRAbMv7bnnk8mXYSC9gTGGtWvX4ssvv8TRo0cxePBgt+UpKSmQy+VucV+6dAmlpaWuuFNTU1FUVOQ2OU12dja0Wi1GjRrlqtPTQ1oeeeQRFBUVoaCgwPU1adIkPPvss67/B8q+tKtXujn6uI6GgfjDCy+8wHQ6HcvNzWXl5eWur4aGBled559/niUkJLCjR4+yM2fOsNTUVJaamupa3tyd/Nhjj7GCggJ25MgRFhUV5bU7efPmzay4uJjt2rWrV7qTW/bmBfq+NKNkatLeMBB/AOD1a+/eva46jY2NbPXq1SwsLIxpNBq2cOFCVl5e7rad69evs7lz5zK1Ws0iIyPZpk2bmM1mc6tz7NgxNmHCBKZQKNiQIUPcvkdPaZ1MgbwvzWgIBiESueevmQiRCiUTIRKhZCJEIpRMhEiEkokQiVAyESIRSiZCJELJRIhEKJkIkQglE5HEQw895Danw72IkokQiVAy9QOiKCIrKwuDBw+GWq3G+PHjcfDgQYiiiEGDBuHDDz90q3/u3DnwPI8bN24AAGpra7Fy5UpERUVBq9Vi9uzZKCwsdNX/3e9+hwkTJuCTTz5BUlISdDodFi9ejLq6OgDAr3/9a+Tl5eH9998Hx3HgOA7Xr1/vtf3vM3rtkVrSY9566y02YsQIduTIEXblyhW2d+9eplQqWW5uLnvppZfYjBkz3Opv2rTJrSwtLY098cQT7PTp0+zHH39kmzZtYhEREay6upoxxti2bdtYcHAwe+qpp1hRURE7fvw4i4mJYa+88gpjzDkhSmpqKlu1apVrqIjdbu+9H0AfQckU4MxmM9NoNOzkyZNu5StWrGBLlixh586dYxzHsRs3bjDGnDMvxcXFsQ8//JAxxti///1vptVqPd4wcd9997GPPvqIMeZMJo1Gw4xGo2v55s2b2dSpU12fWw+puBfRHBAB7vLly2hoaMCjjz7qVm61WjFx4kRMmDABI0eOxKeffoqtW7ciLy8PlZWVWLRoEQCgsLAQJpMJERERbus3NjbiypUrrs9JSUkICQlxfR44cKDbqFdCE6oEPJPJBAA4dOgQ4uLi3JY1v3702WefdSXTp59+ijlz5riSx2QyYeDAgcjNzfXYdsvZguRyudsyjuMgiqKEexL4KJkC3KhRo6BUKlFaWopZs2Z5rfPLX/4Sr776Ks6ePYuDBw9i9+7drmXJycnQ6/WQyWRISkryOQ6FQgGHw+Hz+v0BJVOACwkJwUsvvYSNGzdCFEXMmDEDBoMBJ06cgFarxbJly5CUlITp06djxYoVcDgcePLJJ13rp6WlITU1FQsWLMD27dsxfPhw3Lp1C4cOHcLChQsxadKkTsWRlJSEU6dO4fr16wgODkZ4eDh4/t7qLL639rafevPNN/Haa68hKysLI0eOxJw5c3Do0CG3GY2effZZFBYWYuHChW7TXnEch8OHD2PmzJlYvnw5hg8fjsWLF+PGjRseEzq256WXXoIgCBg1ahSioqJQWloq6T4GApoDghCJUMtEiEQomQiRCCUTIRKhZCJEIpRMhEiEkokQiVAyESIRSiZCJELJRIhEKJkIkQglEyES+f9u50POXzNWdwAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 200x160 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(2,1.6))\n",
    "plt.subplots_adjust(left=0.3, right=1.0, top=0.95, bottom=0.3)\n",
    "plot_curve(accuracies3[:50000], y_bottom=0, y_top=1, label=\"all\")\n",
    "plt.axhline(0.5, color='#999999', alpha=0.5, linestyle='--', zorder=5)\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.xlabel(\"event\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Compare"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure()\n",
    "# plt.subplots_adjust(left=0.3, right=1.0, top=0.95, bottom=0.3)\n",
    "plot_curve(accuracies1[:50000], color='#1F77B4', color2=\"#82B0D2\", y_bottom=0, y_top=1, label=\"LSTM\")\n",
    "plot_curve(accuracies2[:50000], color=\"#2CA02C\", color2=\"#8ECFC9\", y_bottom=0, y_top=1, label=\"Transformer\")\n",
    "plot_curve(accuracies3[:50000], color=\"#FF7F0E\", color2=\"#FFBE7A\", y_bottom=0, y_top=1, label=\"SeL-NAL\")\n",
    "plt.legend()\n",
    "plt.axhline(0.5, color='#999999', alpha=0.5, linestyle='--', zorder=5)\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.xlabel(\"event\")\n",
    "plt.savefig(\"Figures/compare.pdf\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "PyTorchEnv",
   "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": "3.9.19"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}