lec9

README.md

@@ -12,6 +12,7 @@
 | 10.10.2024 | [Feature engineering](./lectures/lec6-feature_engineering.odp) - [в формате pptx](./lectures/lec6-feature_engineering.pptx) - код в [ноутбуке к mlflow](./assets/mlflow/research.ipynb) |
 | 17.10.2024 | [Feature extraction. Настройка гиперпараметров](./lectures/lec7-feature_selection_hyperparams.odp) - [в формате pptx](./lectures/lec7-feature_selection_hyperparams.pptx) - код в [ноутбуке к mlflow](./assets/mlflow/research.ipynb) |
 | 24.10.2024 | [Архитекура сервиса. API](./lectures/lec8-api.odp) - [в формате pptx](./lectures/lec8-api.pptx) |
+| 14.11.2024 | [Создание микросервиса](./lectures/lec9-webserver.pptx) |
 
 
 ## <span style="color:red">Перенос занятий</span>

assets/service/ml_service/.dockerignore

@@ -0,0 +1 @@
+__pycache__

assets/service/ml_service/Dockerfile

@@ -0,0 +1,14 @@
+FROM python:3.11-slim
+
+COPY . /app
+WORKDIR /app
+
+RUN pip install -r requirements.txt
+
+EXPOSE 8000
+VOLUME /models
+
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000" ]
+
+# docker build . --tag estate_model:0
+# docker run -p 8001:8000 -v $(pwd)/../models:/models estate_model:0

assets/service/ml_service/api_handler.py

@@ -0,0 +1,20 @@
+
+import logging
+import pandas as pd
+import pickle as pkl
+
+logger = logging.getLogger("uvicorn.error")
+class FastAPIHandler():
+
+    def __init__(self):
+        logger.warning('Loading model...')
+        try:
+            self.model = pkl.load(open('../models/model.pkl', 'rb'))
+            logger.info('Model is loaded')
+        except Exception as e:
+            logger.error('Error loading model')
+
+    def predict(self, item_features:dict):
+        item_df = pd.DataFrame(data=item_features, index=[0])
+        prediction = self.model.predict(item_df)
+        return (prediction[0])

assets/service/ml_service/main.py

@@ -0,0 +1,19 @@
+
+import random
+from fastapi import FastAPI
+from api_handler import FastAPIHandler
+
+app = FastAPI()
+app.handler = FastAPIHandler()
+
+@app.get('/')
+def root_dir():
+    return({'Hello': 'world'})
+
+@app.post('/api/prediction')
+def make_prediction(flat_id: int, item_features: dict):
+    prediction = app.handler.predict(item_features)
+    return ({
+             'price': prediction,
+             'flat_id': flat_id
+            })

assets/service/ml_service/requirements.txt

@@ -0,0 +1,5 @@
+fastapi
+uvicorn
+pandas
+pickle4
+scikit-learn

assets/service/models/get_model.py

@@ -0,0 +1,18 @@
+import mlflow
+import pickle as pkl
+
+# Работаем с MLflow локально
+TRACKING_SERVER_HOST = "127.0.0.1"
+TRACKING_SERVER_PORT = 5001
+
+registry_uri = f"http://{TRACKING_SERVER_HOST}:{TRACKING_SERVER_PORT}"
+tracking_uri = f"http://{TRACKING_SERVER_HOST}:{TRACKING_SERVER_PORT}"
+
+mlflow.set_tracking_uri(tracking_uri)   
+mlflow.set_registry_uri(registry_uri)   
+
+RUN_NAME = '96a46920d98c48dfa3c019926b44018b'
+loaded_model = mlflow.sklearn.load_model(f'runs:/{RUN_NAME}/models')
+
+with open('model.pkl', 'wb+') as f:
+    pkl.dump(loaded_model, f)