add gradio app + model

app.py

@@ -0,0 +1,125 @@
+# app.py
+import json
+import joblib
+import numpy as np
+import gradio as gr
+from pathlib import Path
+
+PIPE_PATH = Path("best_model/pipeline.joblib")
+META_PATH = Path("best_model/model_meta.json")
+
+# ---- load once ----
+pipeline = joblib.load(PIPE_PATH)
+
+# Try to respect feature order saved in metadata; fallback to a known order
+default_order = [
+    "ph",
+    "Hardness",
+    "Solids",
+    "Chloramines",
+    "Sulfate",
+    "Conductivity",
+    "Organic_carbon",
+    "Trihalomethanes",
+    "Turbidity",
+]
+
+if META_PATH.exists():
+    with open(META_PATH, "r") as f:
+        meta = json.load(f)
+    feature_order = meta.get("feature_order") or meta.get("features_used_in_order") or default_order
+else:
+    feature_order = default_order
+
+label_map = {0: "Unsafe (not potable)", 1: "Safe (potable)"}
+
+def predict_one(
+        ph, Hardness, Solids, Chloramines, Sulfate, Conductivity, Organic_carbon, Trihalomethanes, Turbidity
+):
+    values = {
+        "ph": ph,
+        "Hardness": Hardness,
+        "Solids": Solids,
+        "Chloramines": Chloramines,
+        "Sulfate": Sulfate,
+        "Conductivity": Conductivity,
+        "Organic_carbon": Organic_carbon,
+        "Trihalomethanes": Trihalomethanes,
+        "Turbidity": Turbidity,
+    }
+    X = np.array([[values[k] for k in feature_order]])
+    proba = getattr(pipeline, "predict_proba", None)
+    if proba:
+        p = float(proba(X)[0, 1])
+    else:
+        # if model has no predict_proba, estimate from decision_function or 0/1
+        y_hat = int(pipeline.predict(X)[0])
+        p = 0.5 if y_hat == 1 else 0.0
+    y = int(pipeline.predict(X)[0])
+    return label_map[y], round(p, 4), feature_order
+
+# --- A simple "chat" wrapper that lets users type numbers freely ---
+import re
+def chat_predict(message, history):
+    # Accept formats like: "ph=7.2 Hardness: 160 ...", or a JSON blob.
+    try:
+        if message.strip().startswith("{"):
+            values = json.loads(message)
+        else:
+            # extract key=value pairs crudely
+            pairs = dict(re.findall(r'(\b[a-zA-Z_]+)\s*[:=]\s*([+-]?\d+(\.\d+)?)', message))
+            values = {k: float(v) for k, v in pairs.items()}
+        # Fill missing with NaN -> pipeline imputer will handle
+        x = []
+        for k in feature_order:
+            x.append(float(values.get(k, np.nan)))
+        X = np.array([x])
+        proba = getattr(pipeline, "predict_proba", None)
+        if proba:
+            p = float(proba(X)[0, 1])
+        else:
+            y_hat = int(pipeline.predict(X)[0])
+            p = 0.5 if y_hat == 1 else 0.0
+        y = int(pipeline.predict(X)[0])
+        return f"{label_map[y]} (p={p:.4f})"
+    except Exception as e:
+        return ("Please provide inputs as JSON or 'key=value' pairs for: "
+                f"{', '.join(feature_order)}. Example:\n"
+                "ph=7.2 Hardness=160 Solids=1800 Chloramines=6.5 Sulfate=220 "
+                "Conductivity=350 Organic_carbon=6 Trihalomethanes=25 Turbidity=2.5")
+
+# --------- Gradio UI ----------
+with gr.Blocks(title="Water Potability Predictor") as demo:
+    gr.Markdown("# 💧 Water Potability Predictor\nEnter water chemistry and get a quick potability prediction.")
+    with gr.Tabs():
+        with gr.Tab("Form"):
+            with gr.Row():
+                ph = gr.Number(value=7.0, label="ph")
+                Hardness = gr.Number(value=150.0, label="Hardness (mg/L)")
+                Solids = gr.Number(value=1000.0, label="Solids (ppm)")
+                Chloramines = gr.Number(value=6.0, label="Chloramines (ppm)")
+                Sulfate = gr.Number(value=200.0, label="Sulfate (mg/L)")
+                Conductivity = gr.Number(value=300.0, label="Conductivity (μS/cm)")
+                Organic_carbon = gr.Number(value=6.0, label="Organic carbon (ppm)")
+                Trihalomethanes = gr.Number(value=40.0, label="Trihalomethanes (μg/L)")
+                Turbidity = gr.Number(value=2.0, label="Turbidity (NTU)")
+            btn = gr.Button("Predict")
+            out_label = gr.Textbox(label="Label")
+            out_prob = gr.Number(label="Probability (class=Safe)")
+            out_order = gr.JSON(label="Feature order used")
+            btn.click(
+                predict_one,
+                inputs=[ph, Hardness, Solids, Chloramines, Sulfate, Conductivity, Organic_carbon, Trihalomethanes, Turbidity],
+                outputs=[out_label, out_prob, out_order],
+            )
+        with gr.Tab("Chat"):
+            gr.Markdown(
+                "Type values as JSON or key=value pairs.\n\n"
+                "**Example:** `ph=7.2 Hardness=160 Solids=1800 Chloramines=6.5 "
+                "Sulfate=220 Conductivity=350 Organic_carbon=6 Trihalomethanes=25 Turbidity=2.5`"
+            )
+            chat = gr.ChatInterface(fn=chat_predict, type="messages")
+
+# Spaces will look for this
+if __name__ == "__main__":
+    demo.launch()

best_model/model_meta.json

@@ -0,0 +1,42 @@
+{
+  "artifact": "pipeline.joblib",
+  "created_at": "2025-10-20 10:29:47",
+  "feature_names_in_order": [
+    "ph",
+    "Hardness",
+    "Solids",
+    "Chloramines",
+    "Sulfate",
+    "Conductivity",
+    "Organic_carbon",
+    "Trihalomethanes",
+    "Turbidity"
+  ],
+  "target": "Potability",
+  "winner": "RandomForestClassifier",
+  "cv_best_params": {
+    "clf": "RandomForestClassifier",
+    "clf__max_depth": null,
+    "clf__min_samples_split": 5,
+    "clf__n_estimators": 200
+  },
+  "cv_best_score": 0.649618320610687,
+  "test_accuracy": 0.6432926829268293,
+  "preprocessing": {
+    "KNNImputer_on": [
+      "ph",
+      "Sulfate",
+      "Trihalomethanes"
+    ],
+    "MedianImputer_on": [
+      "Hardness",
+      "Solids",
+      "Chloramines",
+      "Conductivity",
+      "Organic_carbon",
+      "Turbidity"
+    ],
+    "SMOTE": true
+  },
+  "scoring": "accuracy"
+}

best_model/pipeline.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:689cf58d5d0736b3a182aa46f2b840f977c5ebf45b48b1828e05ebf2c1fe2e47
+size 14524206