Update app.py
Browse files
app.py
CHANGED
|
@@ -4,42 +4,28 @@ import pandas as pd
|
|
| 4 |
import numpy as np
|
| 5 |
import traceback
|
| 6 |
|
| 7 |
-
#
|
| 8 |
-
# Must match the file you uploaded to the Space
|
| 9 |
-
BUNDLE_FILENAME = "best_model_random_forest.joblib"
|
| 10 |
|
| 11 |
-
#
|
| 12 |
-
# Expected bundle format: {"model": fitted_pipeline_or_model, "feature_cols": [...]}
|
| 13 |
try:
|
| 14 |
bundle = joblib.load(BUNDLE_FILENAME)
|
| 15 |
model = bundle["model"]
|
| 16 |
-
feature_cols = bundle.get("feature_cols", None)
|
| 17 |
except Exception as e:
|
| 18 |
-
# Fail fast with a clear message in the Space logs
|
| 19 |
raise RuntimeError(f"Failed to load model bundle '{BUNDLE_FILENAME}': {e}")
|
| 20 |
|
| 21 |
-
# ββ 3) OPTIONAL: feature-engineering used during training ββββββββββββββββββββββ
|
| 22 |
-
# If you engineered these in the notebook, keep them here. If not, they are harmless.
|
| 23 |
def add_features(df: pd.DataFrame) -> pd.DataFrame:
|
| 24 |
df = df.copy()
|
| 25 |
-
# Product code prefix from Product_Id (if present)
|
| 26 |
if "Product_Id" in df.columns:
|
| 27 |
df["Product_Category_Code"] = df["Product_Id"].astype(str).str[:2]
|
| 28 |
-
|
| 29 |
-
# Store age from establishment year (if present)
|
| 30 |
if "Store_Establishment_Year" in df.columns:
|
| 31 |
-
# keep same constant year you used in training; adjust if needed
|
| 32 |
df["Store_Age"] = 2025 - pd.to_numeric(df["Store_Establishment_Year"], errors="coerce")
|
| 33 |
-
|
| 34 |
-
# Price per weight (if both present)
|
| 35 |
if ("Product_MRP" in df.columns) and ("Product_Weight" in df.columns):
|
| 36 |
w = pd.to_numeric(df["Product_Weight"], errors="coerce").replace(0, np.nan)
|
| 37 |
p = pd.to_numeric(df["Product_MRP"], errors="coerce")
|
| 38 |
df["Price_per_weight"] = p / w
|
| 39 |
-
|
| 40 |
return df
|
| 41 |
|
| 42 |
-
# ββ 4) FLASK APP βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 43 |
app = Flask(__name__)
|
| 44 |
|
| 45 |
@app.route("/", methods=["GET"])
|
|
@@ -52,7 +38,6 @@ def health():
|
|
| 52 |
|
| 53 |
@app.route("/schema", methods=["GET"])
|
| 54 |
def schema():
|
| 55 |
-
"""Quick way to see which raw columns the backend expects before preprocessing."""
|
| 56 |
return jsonify({
|
| 57 |
"expects_columns_like": [
|
| 58 |
"Product_Id", "Product_Weight", "Product_Sugar_Content",
|
|
@@ -61,22 +46,13 @@ def schema():
|
|
| 61 |
"Store_Location_City_Type", "Store_Type"
|
| 62 |
],
|
| 63 |
"engineers": ["Product_Category_Code", "Store_Age", "Price_per_weight"],
|
| 64 |
-
"model_feature_cols": feature_cols
|
| 65 |
}), 200
|
| 66 |
|
| 67 |
@app.route("/predict", methods=["POST"])
|
| 68 |
def predict():
|
| 69 |
-
"""
|
| 70 |
-
Accepts:
|
| 71 |
-
- a single JSON object, or
|
| 72 |
-
- a list of JSON objects.
|
| 73 |
-
Returns:
|
| 74 |
-
{"predictions": [float, ...]}
|
| 75 |
-
"""
|
| 76 |
try:
|
| 77 |
payload = request.get_json(force=True)
|
| 78 |
-
|
| 79 |
-
# Accept dict or list[dict]
|
| 80 |
if isinstance(payload, dict):
|
| 81 |
df = pd.DataFrame([payload])
|
| 82 |
elif isinstance(payload, list):
|
|
@@ -84,31 +60,22 @@ def predict():
|
|
| 84 |
else:
|
| 85 |
return jsonify({"error": "Input must be a JSON object or a list of objects."}), 400
|
| 86 |
|
| 87 |
-
# 1) Apply the same feature engineering used in training
|
| 88 |
df = add_features(df)
|
| 89 |
|
| 90 |
-
# 2) If you saved a feature list, align to it. Let the model/pipeline handle imputation/encoding.
|
| 91 |
if feature_cols is not None:
|
| 92 |
-
# Keep only the columns the model was trained on; add any missing as NaN
|
| 93 |
for c in feature_cols:
|
| 94 |
if c not in df.columns:
|
| 95 |
df[c] = np.nan
|
| 96 |
X = df[feature_cols]
|
| 97 |
else:
|
| 98 |
-
# No explicit feature list in bundle; pass raw (the pipeline inside model should handle it)
|
| 99 |
X = df
|
| 100 |
|
| 101 |
-
# 3) Predict
|
| 102 |
preds = model.predict(X)
|
| 103 |
preds = [float(p) for p in np.ravel(preds)]
|
| 104 |
-
|
| 105 |
return jsonify({"predictions": preds}), 200
|
| 106 |
|
| 107 |
except Exception as e:
|
| 108 |
-
# Include traceback for fast debugging from the Space logs
|
| 109 |
return jsonify({"error": str(e), "traceback": traceback.format_exc()}), 400
|
| 110 |
|
| 111 |
-
|
| 112 |
if __name__ == "__main__":
|
| 113 |
-
# For local docker run; in Spaces we use gunicorn (see your Dockerfile)
|
| 114 |
app.run(host="0.0.0.0", port=5000)
|
|
|
|
| 4 |
import numpy as np
|
| 5 |
import traceback
|
| 6 |
|
| 7 |
+
BUNDLE_FILENAME = "best_model_random_forest.joblib" # must match the file in the repo
|
|
|
|
|
|
|
| 8 |
|
| 9 |
+
# Load bundle: {"model": fitted_pipeline_or_model, "feature_cols": [...]}
|
|
|
|
| 10 |
try:
|
| 11 |
bundle = joblib.load(BUNDLE_FILENAME)
|
| 12 |
model = bundle["model"]
|
| 13 |
+
feature_cols = bundle.get("feature_cols", None)
|
| 14 |
except Exception as e:
|
|
|
|
| 15 |
raise RuntimeError(f"Failed to load model bundle '{BUNDLE_FILENAME}': {e}")
|
| 16 |
|
|
|
|
|
|
|
| 17 |
def add_features(df: pd.DataFrame) -> pd.DataFrame:
|
| 18 |
df = df.copy()
|
|
|
|
| 19 |
if "Product_Id" in df.columns:
|
| 20 |
df["Product_Category_Code"] = df["Product_Id"].astype(str).str[:2]
|
|
|
|
|
|
|
| 21 |
if "Store_Establishment_Year" in df.columns:
|
|
|
|
| 22 |
df["Store_Age"] = 2025 - pd.to_numeric(df["Store_Establishment_Year"], errors="coerce")
|
|
|
|
|
|
|
| 23 |
if ("Product_MRP" in df.columns) and ("Product_Weight" in df.columns):
|
| 24 |
w = pd.to_numeric(df["Product_Weight"], errors="coerce").replace(0, np.nan)
|
| 25 |
p = pd.to_numeric(df["Product_MRP"], errors="coerce")
|
| 26 |
df["Price_per_weight"] = p / w
|
|
|
|
| 27 |
return df
|
| 28 |
|
|
|
|
| 29 |
app = Flask(__name__)
|
| 30 |
|
| 31 |
@app.route("/", methods=["GET"])
|
|
|
|
| 38 |
|
| 39 |
@app.route("/schema", methods=["GET"])
|
| 40 |
def schema():
|
|
|
|
| 41 |
return jsonify({
|
| 42 |
"expects_columns_like": [
|
| 43 |
"Product_Id", "Product_Weight", "Product_Sugar_Content",
|
|
|
|
| 46 |
"Store_Location_City_Type", "Store_Type"
|
| 47 |
],
|
| 48 |
"engineers": ["Product_Category_Code", "Store_Age", "Price_per_weight"],
|
| 49 |
+
"model_feature_cols": feature_cols
|
| 50 |
}), 200
|
| 51 |
|
| 52 |
@app.route("/predict", methods=["POST"])
|
| 53 |
def predict():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 54 |
try:
|
| 55 |
payload = request.get_json(force=True)
|
|
|
|
|
|
|
| 56 |
if isinstance(payload, dict):
|
| 57 |
df = pd.DataFrame([payload])
|
| 58 |
elif isinstance(payload, list):
|
|
|
|
| 60 |
else:
|
| 61 |
return jsonify({"error": "Input must be a JSON object or a list of objects."}), 400
|
| 62 |
|
|
|
|
| 63 |
df = add_features(df)
|
| 64 |
|
|
|
|
| 65 |
if feature_cols is not None:
|
|
|
|
| 66 |
for c in feature_cols:
|
| 67 |
if c not in df.columns:
|
| 68 |
df[c] = np.nan
|
| 69 |
X = df[feature_cols]
|
| 70 |
else:
|
|
|
|
| 71 |
X = df
|
| 72 |
|
|
|
|
| 73 |
preds = model.predict(X)
|
| 74 |
preds = [float(p) for p in np.ravel(preds)]
|
|
|
|
| 75 |
return jsonify({"predictions": preds}), 200
|
| 76 |
|
| 77 |
except Exception as e:
|
|
|
|
| 78 |
return jsonify({"error": str(e), "traceback": traceback.format_exc()}), 400
|
| 79 |
|
|
|
|
| 80 |
if __name__ == "__main__":
|
|
|
|
| 81 |
app.run(host="0.0.0.0", port=5000)
|