Create dashboard.py

dashboard.py

@@ -0,0 +1,74 @@
+# dashboard.py
+
+import firebase_admin
+from firebase_admin import credentials, firestore
+import pandas as pd
+import matplotlib.pyplot as plt
+import io
+import gradio as gr
+
+# Initialize Firebase Admin SDK (ensure the JSON file is in your space)
+if not firebase_admin._apps:
+    cred = credentials.Certificate("evotransformer-firebase-adminsdk-fbsvc-37a4b838aa.json")
+    firebase_admin.initialize_app(cred)
+
+db = firestore.client()
+
+def fetch_logs():
+    docs = db.collection("evo_feedback_logs").stream()
+    data = []
+    for doc in docs:
+        d = doc.to_dict()
+        if all(k in d for k in ["goal", "evo", "gpt", "correct"]):
+            data.append(d)
+    return pd.DataFrame(data)
+
+def generate_dashboard():
+    df = fetch_logs()
+    if df.empty:
+        return "No data yet.", None
+
+    df["evo_correct"] = df["evo"] == df["correct"]
+    df["gpt_correct"] = df["gpt"] == df["correct"]
+
+    evo_acc = df["evo_correct"].mean()
+    gpt_acc = df["gpt_correct"].mean()
+    agreement = (df["evo"] == df["gpt"]).mean()
+
+    fig, ax = plt.subplots(figsize=(6, 4))
+    counts = [
+        df["evo_correct"].sum(),
+        df["evo_correct"].count() - df["evo_correct"].sum(),
+        df["gpt_correct"].sum(),
+        df["gpt_correct"].count() - df["gpt_correct"].sum()
+    ]
+    bars = ax.bar(["Evo Correct", "Evo Wrong", "GPT Correct", "GPT Wrong"], counts)
+    ax.set_title("Model Accuracy Summary")
+    ax.set_ylabel("Count")
+
+    buf = io.BytesIO()
+    plt.tight_layout()
+    plt.savefig(buf, format="png")
+    plt.close(fig)
+    buf.seek(0)
+
+    summary = (
+        f"🧠 Evo Accuracy: {evo_acc:.2%}\n"
+        f"🤖 GPT-3.5 Accuracy: {gpt_acc:.2%}\n"
+        f"⚖️ Agreement Rate: {agreement:.2%}\n"
+        f"📝 Total Examples Logged: {len(df)}"
+    )
+
+    return summary, buf
+
+with gr.Blocks() as dashboard:
+    gr.Markdown("## 📊 EvoTransformer Live Evolution Dashboard")
+    gr.Markdown("> 🔄 *Below metrics are based on real-time user feedback.*")
+    
+    summary_text = gr.Textbox(label="Summary", interactive=False)
+    plot = gr.Image(type="pil", label="Accuracy Plot")
+
+    refresh = gr.Button("🔁 Refresh Data")
+    refresh.click(fn=generate_dashboard, outputs=[summary_text, plot])
+
+dashboard.launch()