Update app.py

app.py

@@ -1,140 +1,108 @@
 import gradio as gr
-import numpy as np
 import pandas as pd
-import torch
+import numpy as np
 import time
-from transformers import AutoTokenizer, AutoModelForCausalLM
-
-# ---------- Model ----------
-MODEL_NAME = "microsoft/Phi-3-mini-4k-instruct"
+from transformers import pipeline
 
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_NAME,
-    device_map="cpu",
-    torch_dtype=torch.float32
-)
+# ---- LLM ----
+MODEL_NAME = "google/flan-t5-base"
+llm = pipeline("text2text-generation", model=MODEL_NAME)
 
-# ---------- Trend Logic ----------
+# ---- Helpers ----
 def detect_trend(values):
-    diffs = np.diff(values)
-    if len(diffs) == 0:
-        return "FLAT"
-    if all(d > 0 for d in diffs):
+    if len(values) < 2:
+        return "NO DATA"
+    x = np.arange(len(values))
+    y = np.array(values)
+    slope = np.polyfit(x, y, 1)[0]
+    if slope > 0:
         return "INCREASING"
-    elif all(d < 0 for d in diffs):
+    elif slope < 0:
         return "DECREASING"
-    else:
-        return "MIXED"
-
+    return "STABLE"
 
 def change_score(values):
-    values = np.array(values, dtype=float)
     if len(values) < 2:
         return 0
-    std = np.std(values)
-    change = abs(values[-1] - values[0])
-    return round((change / (std + 1e-6)), 4)
+    return float(np.std(values))
 
-# ---------- LLM Explanation ----------
 def explain(kpi, values, trend):
-    # ✅ limit values to max 50 numbers
-    short_values = values[:50]
-
+    values = values[:50]  # prevent token overflow
     prompt = f"""
-    You are a telecom KPI expert.
-    KPI name: {kpi}
-    Values (sample): {short_values}
-    Detected trend: {trend}
-    
-    In 2 short sentences explain what this trend might mean.
+    Explain KPI trend.
+    KPI: {kpi}
+    Trend: {trend}
+    Values: {values}
+    Explain shortly.
     """
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
-    output = model.generate(**inputs, max_new_tokens=120)
-    return tokenizer.decode(output[0], skip_special_tokens=True)
+    out = llm(prompt, max_new_tokens=80)
+    return out[0]["generated_text"]
+
+# ---- MAIN FUNCTION ----
+def analyze(file, progress=gr.Progress()):
+    start_time = time.time()
+    print("✅ Analyze function triggered")
 
-# ---------- File Loader ----------
-def load_csv(file):
     if file is None:
-        return None, gr.update(choices=[])
+        return "❌ Upload a CSV file"
 
     df = pd.read_csv(file.name)
-    numeric_cols = df.select_dtypes(include=np.number).columns.tolist()
-
-    return df, gr.update(choices=numeric_cols, value=numeric_cols[:5])
 
-# ---------- Main Analysis ----------
-def analyze(df, selected_kpis, progress=gr.Progress()):
-    start = time.time()
-    print("✅ Analyze function triggered")
-
-    if df is None or not selected_kpis:
-        return "❌ Upload CSV and select KPI columns"
+    # Auto-detect numeric KPIs
+    kpis = df.select_dtypes(include=np.number).columns.tolist()
 
-    progress(0, desc="Starting analysis...")
+    if not kpis:
+        return "❌ No numeric KPI columns found"
 
     results = []
-    explanations = []
-    total = len(selected_kpis)
+    total = len(kpis)
 
-    for i, kpi in enumerate(selected_kpis):
-        progress((i + 1) / total, desc=f"Analyzing {kpi}...")
+    # ---- Spinner + Progress ----
+    for i, kpi in enumerate(kpis):
+        progress((i+1)/total, desc=f"Analyzing {kpi}...")
 
-        vals = df[kpi].dropna().values.tolist()
+        vals = df[kpi].dropna().tolist()
         trend = detect_trend(vals)
         score = change_score(vals)
 
         results.append((kpi, trend, score, vals))
 
+    # Rank KPIs
     ranked = sorted(results, key=lambda x: x[2], reverse=True)
 
-    # LLM step
+    # ---- LLM explanations ----
     progress(0.9, desc="Generating LLM explanations...")
 
+    explanations = []
     for kpi, trend, score, vals in ranked[:5]:
-        short_vals = vals[:50]   # ✅ FIX TOKEN OVERFLOW
         try:
-            exp = explain(kpi, short_vals, trend)
+            exp = explain(kpi, vals, trend)
         except Exception as e:
-            exp = f"⚠️ LLM Error: {str(e)}"
+            exp = f"LLM error: {e}"
         explanations.append(f"**{kpi}** → {exp}")
 
-    progress(1.0, desc="Done ✅")
-
-    elapsed = round(time.time() - start, 2)
-
-    text = f"⏱ Time taken: {elapsed}s\n\n"
-    text += "## 🔥 Top 5 KPIs\n"
+    elapsed = round(time.time() - start_time, 2)
 
+    # ---- Final Output ----
+    output = f"⏱ Time taken: {elapsed} sec\n\n"
+    output += "🔥 Top 5 KPIs with most change:\n\n"
     for kpi, trend, score, _ in ranked[:5]:
-        text += f"**{kpi}** → {trend} (Score: {score})\n\n"
+        output += f"- {kpi}: {trend} (σ={round(score,2)})\n"
 
-    text += "## 🧠 LLM Explanations\n"
+    output += "\n🧠 LLM Insights:\n\n"
     for e in explanations:
-        text += f"{e}\n\n"
-
-    return text
-
-# ---------- UI ----------
-with gr.Blocks() as demo:
-    gr.Markdown("# 🤖 KPI Trend Analyzer with LLM Explanation")
-
-    file = gr.File(label="Upload CSV")
-    kpi_select = gr.CheckboxGroup(label="Select KPI columns", choices=[])
+        output += e + "\n\n"
 
-    df_state = gr.State()
+    return output
 
-    file.upload(fn=load_csv, inputs=file, outputs=[df_state, kpi_select])
+# ---- UI ----
+with gr.Blocks() as app:
+    gr.Markdown("## 📊 KPI Trend Analyzer (LLM Powered)")
 
-    analyze_btn = gr.Button("Analyze")
-    output = gr.Markdown()
+    file_input = gr.File(label="Upload CSV")
+    analyze_btn = gr.Button("Analyze KPIs")
+    output_box = gr.Markdown()
 
-    analyze_btn.click(
-        fn=analyze,
-        inputs=[df_state, kpi_select],
-        outputs=output,
-        show_progress=True
-    )
+    analyze_btn.click(fn=analyze, inputs=[file_input], outputs=[output_box])
 
-# ✅ Share link required in HF Spaces
-demo.launch(share=True)
+app.launch()  # ⚠️ no share=True