Remove local models, load from HF Model Hub

Dockerfile

@@ -6,7 +6,6 @@ COPY requirements_api.txt .
 RUN pip install --no-cache-dir -r requirements_api.txt
 
 COPY src src
-COPY models models
 COPY data data
 
 EXPOSE 7860

requirements.txt

@@ -6,3 +6,4 @@ scikit-learn
 joblib
 nltk
 prometheus-client
+huggingface-hub

src/api.py

@@ -1,9 +1,10 @@
-from fastapi import FastAPI, Request
+from fastapi import FastAPI
 from pydantic import BaseModel
-import joblib
 import numpy as np
 import pandas as pd
 import time
+import joblib
+from huggingface_hub import hf_hub_download
 
 from prometheus_client import (
     Counter,
@@ -13,14 +14,14 @@ from prometheus_client import (
 )
 from fastapi.responses import Response
 
-# -----------------------------
+# =====================================================
 # App
-# -----------------------------
+# =====================================================
 app = FastAPI(title="Investor Sentiment Inference API")
 
-# -----------------------------
-# Prometheus metrics
-# -----------------------------
+# =====================================================
+# Prometheus Metrics
+# =====================================================
 REQUEST_COUNT = Counter(
     "prediction_requests_total",
     "Total number of prediction requests"
@@ -37,18 +38,39 @@ SENTIMENT_DISTRIBUTION = Histogram(
     buckets=(-1, -0.5, 0, 0.5, 1)
 )
 
-# -----------------------------
-# Load model + scaler
-# -----------------------------
-MODEL_PATH = "models/AAPL"
-model = joblib.load(f"{MODEL_PATH}/rf.joblib")
-scaler_x = joblib.load(f"{MODEL_PATH}/scaler_x.joblib")
-
-# -----------------------------
-# Sentiment logic
-# -----------------------------
-POS_WORDS = {"good", "buy", "up", "rise", "gain", "bull", "profit", "growth", "bullish"}
-NEG_WORDS = {"bad", "sell", "down", "fall", "loss", "bear", "risk", "crash", "bearish"}
+# =====================================================
+# Load model dynamically from Hugging Face Hub
+# =====================================================
+HF_MODEL_REPO = "Mayur-cinderace/investormlops-models"
+TICKER = "AAPL"
+
+def load_model():
+    model_path = hf_hub_download(
+        repo_id=HF_MODEL_REPO,
+        filename=f"{TICKER}/rf.joblib"
+    )
+    scaler_path = hf_hub_download(
+        repo_id=HF_MODEL_REPO,
+        filename=f"{TICKER}/scaler_x.joblib"
+    )
+
+    model = joblib.load(model_path)
+    scaler = joblib.load(scaler_path)
+    return model, scaler
+
+model, scaler_x = load_model()
+
+# =====================================================
+# Sentiment Logic
+# =====================================================
+POS_WORDS = {
+    "good", "buy", "up", "rise", "gain", "bull",
+    "profit", "growth", "bullish", "strong"
+}
+NEG_WORDS = {
+    "bad", "sell", "down", "fall", "loss",
+    "bear", "risk", "crash", "bearish", "weak"
+}
 
 def simple_sentiment(text: str) -> float:
     words = text.lower().split()
@@ -56,23 +78,23 @@ def simple_sentiment(text: str) -> float:
     neg = sum(w in NEG_WORDS for w in words)
     return (pos - neg) / (pos + neg) if (pos + neg) > 0 else 0.0
 
-# -----------------------------
-# Input schema
-# -----------------------------
+# =====================================================
+# Input Schema
+# =====================================================
 class InputText(BaseModel):
     sentence: str
 
-# -----------------------------
-# Market context
-# -----------------------------
+# =====================================================
+# Market Context (latest available features)
+# =====================================================
 def get_latest_market_context():
     df = pd.read_csv("data/processed/merged_features.csv")
-    last = df[df["Ticker"] == "AAPL"].iloc[-1]
+    last = df[df["Ticker"] == TICKER].iloc[-1]
     return last["return_lag1"], last["volume_lag1"]
 
-# -----------------------------
-# Prediction endpoint
-# -----------------------------
+# =====================================================
+# Prediction Endpoint
+# =====================================================
 @app.post("/predict")
 def predict(data: InputText):
     start_time = time.time()
@@ -85,26 +107,26 @@ def predict(data: InputText):
 
     X = np.array([[return_lag1, volume_lag1, sentiment]])
     Xs = scaler_x.transform(X)
-    pred = model.predict(Xs)[0]
+    prediction = model.predict(Xs)[0]
 
     REQUEST_LATENCY.observe(time.time() - start_time)
 
     return {
         "sentence": data.sentence,
         "sentiment_score": sentiment,
-        "predicted_return": float(pred)
+        "predicted_return": float(prediction)
     }
 
-# -----------------------------
-# Prometheus scrape endpoint
-# -----------------------------
+# =====================================================
+# Prometheus Metrics Endpoint
+# =====================================================
 @app.get("/metrics")
 def metrics():
     return Response(generate_latest(), media_type=CONTENT_TYPE_LATEST)
 
-# -----------------------------
-# Health check (very important)
-# -----------------------------
+# =====================================================
+# Health Check
+# =====================================================
 @app.get("/health")
 def health():
     return {"status": "ok"}