mlops fashion mnist automatic template (#232)

* new template

* update code to properly use .file api path

* removed local example file

templates/python39-mlops-fashionmnist/algorithmia.conf

@@ -0,0 +1,5 @@
+{
+    "username": "__USER__",
+    "algoname": "__ALGO__",
+    "langauge": "python3.9"
+}

templates/python39-mlops-fashionmnist/requirements.txt

@@ -0,0 +1,7 @@
+algorithmia>=1.0.0,<2.0
+datarobot-mlops==8.0.7
+pyaml==21.10.1
+pillow<9.0
+torch==1.11.0
+torchvision==0.12.0
+numpy>=1.21

templates/python39-mlops-fashionmnist/src/__ALGO__.py

@@ -0,0 +1,42 @@
+from Algorithmia import ADK
+from datarobot.mlops.mlops import MLOps
+import torch
+from torchvision import transforms
+from time import time
+import pandas as pd
+from PIL import Image
+from src.labels import labels_map
+import os
+
+# The model itself is stored in an Algorithmia Data Collection
+def load(state):
+    # We load the model in this way to ensure that the model is loaded only once
+    model_path = state.client.file(os.environ['MODEL_PATH']).getFile(as_path=True)
+    state['model'] = torch.jit.load(model_path)
+    state['mlops'] = MLOps().init()
+    state['labels'] = labels_map
+    return state
+
+
+# As we're using the ADK system, state is a dictionary that contains the model and the labels
+# Input is expected to be an image file on the Algorithmia Data API; more information on the Data API can be found here.
+# https://algorithmia.com/developers/data/hosted
+def apply(input, state):
+    start_t = time()
+    local_img = state.client.file(input).getFile(as_path=True)
+    transform = transforms.Compose([transforms.Resize(28),
+                                    transforms.ToTensor()])
+    img = Image.open(local_img).convert('L')
+    tensor = transform(img)
+    output = state['model'].forward(tensor)
+    _, predicted = torch.max(output.data, 1)
+    output = torch.softmax(output, 0).reshape(1, -1)
+    prediction = state['labels'][int(predicted.item())]
+    end_t = time()
+    state['mlops'].report_deployment_stats(1, end_t - start_t)
+    state['mlops'].report_predictions_data(predictions=output.tolist())
+    return prediction
+
+
+algorithm = ADK(apply, load)
+algorithm.init(mlops=True)