Ml inference in OS

docker-compose.yaml

@@ -34,6 +34,24 @@ services:
     networks:
       - opensearch-net
 
+  inference:
+    container_name: inference-api
+    build:
+      context: ./inference_api
+      dockerfile: Dockerfile
+    ports:
+      - "8000:8000"
+    environment:
+      - OMP_NUM_THREADS=1
+    deploy:
+      resources:
+        limits:
+          memory: 1G
+        reservations:
+          memory: 512M
+    networks:
+      - opensearch-net
+
 networks:
   opensearch-net:
     driver: bridge

inference_api/Dockerfile

@@ -0,0 +1,15 @@
+FROM python:3.12.6
+
+COPY ./requirements.txt /code/requirements.txt
+COPY ./model/regression_model.pkl /code/model/regression_model.pkl
+#
+RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
+COPY ./app /code/app
+
+# Run the API
+WORKDIR /code/app
+
+#RUN fastapi run opensearch_inference_service.py
+CMD ["fastapi", "run", "opensearch_inference_service.py"]
+#CMD ["gunicorn", "main:app", "-b", "0.0.0.0:8000", "--timeout", "600", "-w", "1", "-k", "uvicorn.workers.UvicornWorker"]
+EXPOSE 8000

inference_api/app/opensearch_inference_service.py

@@ -0,0 +1,46 @@
+from fastapi import FastAPI
+from pydantic import BaseModel
+import pickle
+import pandas as pd
+
+# Create a FastAPI application instance
+app = FastAPI()
+
+# Load the pre-trained machine learning model
+with open('/code/model/regression_model.pkl', 'rb') as fid:
+    model = pickle.load(fid)
+
+# Define a Pydantic model for input validation
+class InputData(BaseModel):
+    data: str
+
+# Define a POST endpoint for making predictions
+@app.post("/predict/")
+def predict(input_data: InputData):
+    """
+    Accept a JSON object with a key "data" containing a comma-separated string of floats.
+    Convert the string to a list of floats and use the model to make predictions.
+    """
+    try:
+        # Split the input string by commas and convert to a list of floats
+        data_list = [float(value.strip()) for value in input_data.data.split(",")]
+
+        # Define the column names for the input features
+        columns = model.feature_names_in_
+
+        # Create a pandas DataFrame from the input data
+        features = pd.DataFrame([data_list], columns=columns)
+
+        # Use the model to make a prediction
+        prediction = round(model.predict(features)[0], 2)
+        keywordness = 1 - prediction
+
+        # Return the prediction as a JSON object
+        return {
+            "neuralness": prediction,
+            "keywordness": keywordness
+        }
+
+    except Exception as e:
+        # Handle errors and provide feedback
+        return {"error": str(e)}

inference_api/requirements.txt

@@ -0,0 +1,3 @@
+fastapi[standard]==0.115.5
+pandas==2.2.3
+scikit-learn==1.5.2