Update app.py

app.py

@@ -11,7 +11,7 @@ from sklearn.ensemble import RandomForestRegressor
 from sklearn.metrics import mean_absolute_error, r2_score
 from sklearn.model_selection import train_test_split
 
-# --- CONFIG ---
+# --- CONFIG & LOGGING ---
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
 
@@ -38,63 +38,66 @@ REPO_CONFIG = {
     }
 }
 
-NON_FEATURE_COLS = {"sample_id", "sample_seed", "circuit_hash", "split", "circuit_qasm", "qasm_raw", "qasm_transpiled"}
+# Колонки, которые нельзя использовать как фичи для обучения
+NON_FEATURE_COLS = {
+    "sample_id", "sample_seed", "circuit_hash", "split", "circuit_qasm", 
+    "qasm_raw", "qasm_transpiled", "circuit_type_resolved", "circuit_type_requested",
+    "noise_type", "noise_prob", "observable_bases", "observable_mode", "backend_device", 
+    "precision_mode", "circuit_signature"
+}
 
 _ASSET_CACHE = {}
 
 def load_all_assets(key: str) -> Dict:
     if key not in _ASSET_CACHE:
+        logger.info(f"Fetching {key}...")
         ds = load_dataset(REPO_CONFIG[key]["repo"])
         meta = requests.get(REPO_CONFIG[key]["meta_url"]).json()
         report = requests.get(REPO_CONFIG[key]["report_url"]).json()
         _ASSET_CACHE[key] = {"df": pd.DataFrame(ds["train"]), "meta": meta, "report": report}
     return _ASSET_CACHE[key]
 
-# --- RENDER FUNCTIONS ---
+# --- UI LOGIC ---
+
 def get_methodology_content(ds_name: str):
     assets = load_all_assets(ds_name)
-    meta = assets["meta"]
+    meta, report = assets["meta"], assets["report"]
     params = meta.get("parameters", {})
-    report = assets["report"]
     
     families = report.get("families", {})
-    # Важно: Markdown таблицы требуют отсутствия пробелов в начале строки
     fam_table = "| Family | Samples | Description |\n|:---|:---|:---|\n"
     for f, count in families.items():
         fam_table += f"| {f.upper()} | {count} | Synthetic {f} circuits |\n"
 
     return f"""
-## 📖 Methodology & Release Notes: {meta.get('dataset_version')}
-
-### 1. Generation Profile
-Dataset produced via **QSBench v{meta.get('generator_version')}**.
-- **Hardware Profile:** {params.get('n_qubits')} Qubits | Depth: {params.get('depth')}
-- **Noise Configuration:** `{params.get('noise')}` (p={params.get('noise_prob')})
-- **Backend:** {meta.get('backend_device')} | {meta.get('precision_mode')} precision
+## 📖 Methodology: {meta.get('dataset_version')}
+**Generator:** QSBench v{meta.get('generator_version')}
+**Config:** {params.get('n_qubits')} Qubits | Depth {params.get('depth')} | Noise `{params.get('noise')}` (p={params.get('noise_prob')})
 
-### 2. Circuit Family Coverage
+### Circuit Family Coverage
 {fam_table}
-
-### 3. Structural Metric Definitions
-- **Gate Entropy:** Measures circuit complexity and gate distribution diversity.
-- **Meyer-Wallach:** Scalar measure of global entanglement.
-- **Adjacency:** Graph density of the qubit interaction map.
     """
 
-def update_explorer(ds_name: str, split_name: str):
+def sync_ml_metrics(ds_name: str):
+    """Динамически находит все доступные числовые метрики для конкретного датасета"""
     assets = load_all_assets(ds_name)
     df = assets["df"]
-    splits = df["split"].unique().tolist() if "split" in df.columns else ["train"]
-    display_df = df[df["split"] == split_name].head(10) if "split" in df.columns else df.head(10)
     
-    raw = display_df["qasm_raw"].iloc[0] if "qasm_raw" in display_df.columns else "// N/A"
-    tr = display_df["qasm_transpiled"].iloc[0] if "qasm_transpiled" in display_df.columns else "// N/A"
-    meta_text = f"### 📋 {ds_name} | Version: {assets['meta'].get('dataset_version')}"
+    # Берем только числа, исключая таргеты и служебные поля
+    numeric_cols = df.select_dtypes(include=[np.number]).columns.tolist()
+    valid_features = [
+        c for c in numeric_cols 
+        if c not in NON_FEATURE_COLS 
+        and not any(x in c for x in ["ideal_", "noisy_", "error_", "sign_"])
+    ]
     
-    return gr.update(choices=splits), display_df, raw, tr, meta_text
+    # Выбираем "золотой стандарт" по умолчанию, если они есть
+    defaults = [f for f in ["gate_entropy", "meyer_wallach", "n_qubits", "depth", "total_gates"] if f in valid_features]
+    
+    return gr.update(choices=valid_features, value=defaults or valid_features[:5])
 
 def train_model(ds_name: str, features: List[str]):
-    if not features: return None, "### ❌ Select features"
+    if not features: return None, "### ❌ Error: No metrics selected."
     assets = load_all_assets(ds_name)
     df = assets["df"]
     target = "ideal_expval_Z_global"
@@ -111,7 +114,7 @@ def train_model(ds_name: str, features: List[str]):
     
     axes[0].scatter(y_test, preds, alpha=0.3, color='#2c3e50')
     axes[0].plot([y.min(), y.max()], [y.min(), y.max()], 'r--', lw=2)
-    axes[0].set_title(f"R² Score: {r2_score(y_test, preds):.3f}")
+    axes[0].set_title(f"Accuracy (R²: {r2_score(y_test, preds):.3f})")
 
     imp = model.feature_importances_
     idx = np.argsort(imp)[-10:]
@@ -119,59 +122,73 @@ def train_model(ds_name: str, features: List[str]):
     axes[1].set_title("Top Metrics Importance")
 
     sns.histplot(y_test - preds, kde=True, ax=axes[2], color='#d35400')
-    axes[2].set_title("Residuals Distribution")
+    axes[2].set_title("Residuals")
 
     plt.tight_layout(pad=3.0)
     return fig, f"**MAE:** {mean_absolute_error(y_test, preds):.4f}"
 
-# --- UI INTERFACE ---
+def update_explorer(ds_name: str, split_name: str):
+    assets = load_all_assets(ds_name)
+    df = assets["df"]
+    splits = df["split"].unique().tolist() if "split" in df.columns else ["train"]
+    display_df = df[df["split"] == split_name].head(10) if "split" in df.columns else df.head(10)
+    
+    raw = display_df["qasm_raw"].iloc[0] if "qasm_raw" in display_df.columns else "// N/A"
+    tr = display_df["qasm_transpiled"].iloc[0] if "qasm_transpiled" in display_df.columns else "// N/A"
+    
+    return gr.update(choices=splits), display_df, raw, tr, f"### 📋 {ds_name} Explorer"
+
+# --- INTERFACE ---
 with gr.Blocks(theme=gr.themes.Soft(), title="QSBench Hub") as demo:
     gr.Markdown("# 🌌 QSBench: Quantum Analytics Hub")
     
     with gr.Tabs():
-        # TAB 1: EXPLORER
         with gr.TabItem("🔎 Explorer"):
-            metadata_info = gr.Markdown("### Loading...")
+            meta_txt = gr.Markdown("### Loading...")
             with gr.Row():
-                ds_select = gr.Dropdown(list(REPO_CONFIG.keys()), value="Core (Clean)", label="Dataset Pack")
-                split_select = gr.Dropdown(["train"], value="train", label="Subset")
-            data_table = gr.Dataframe(interactive=False)
+                ds_sel = gr.Dropdown(list(REPO_CONFIG.keys()), value="Core (Clean)", label="Dataset")
+                sp_sel = gr.Dropdown(["train"], value="train", label="Split")
+            data_view = gr.Dataframe(interactive=False)
             with gr.Row():
-                code_raw = gr.Code(label="Source QASM", language="python")
-                code_tr = gr.Code(label="Transpiled QASM", language="python")
+                c_raw = gr.Code(label="Source QASM", language="python")
+                c_tr = gr.Code(label="Transpiled QASM", language="python")
 
-        # TAB 2: ML
         with gr.TabItem("🤖 ML Training"):
             with gr.Row():
                 with gr.Column(scale=1):
-                    ml_ds = gr.Dropdown(list(REPO_CONFIG.keys()), value="Core (Clean)", label="Select Dataset")
-                    ml_feat = gr.CheckboxGroup(label="Metrics", choices=["gate_entropy", "meyer_wallach", "n_qubits", "depth", "total_gates"], value=["gate_entropy", "meyer_wallach"])
-                    btn = gr.Button("Run Training", variant="primary")
+                    ml_ds_sel = gr.Dropdown(list(REPO_CONFIG.keys()), value="Core (Clean)", label="Select Dataset")
+                    # Динамический список метрик
+                    ml_feat_sel = gr.CheckboxGroup(label="Available Metrics (Auto-detected)", choices=[])
+                    train_btn = gr.Button("Execute Baseline", variant="primary")
                 with gr.Column(scale=2):
-                    plot_out = gr.Plot()
-                    txt_out = gr.Markdown()
+                    p_out = gr.Plot()
+                    t_out = gr.Markdown()
 
-        # TAB 3: METHODOLOGY (С ВЫБОРОМ)
         with gr.TabItem("📖 Methodology"):
-            method_ds_select = gr.Dropdown(list(REPO_CONFIG.keys()), value="Core (Clean)", label="View methodology for:")
-            guide_md = gr.Markdown()
+            meth_ds_sel = gr.Dropdown(list(REPO_CONFIG.keys()), value="Core (Clean)", label="Dataset Details")
+            meth_md = gr.Markdown()
 
-    # LINKS FOOTER
     gr.Markdown(f"""
 ---
 ### 🔗 Project Links
-[**🤗 Hugging Face**](https://huggingface.co/QSBench) | [**💻 GitHub**](https://github.com/QSBench) | [**🌐 Website**](https://qsbench.github.io)
+[**🌐 Website**](https://qsbench.github.io) | [**🤗 Hugging Face**](https://huggingface.co/QSBench) | [**💻 GitHub**](https://github.com/QSBench)
     """)
 
-    # EVENT HANDLERS
-    ds_select.change(update_explorer, [ds_select, split_select], [split_select, data_table, code_raw, code_tr, metadata_info])
-    ml_ds.change(lambda x: gr.update(), [ml_ds], []) # Simple sync
-    method_ds_select.change(get_methodology_content, [method_ds_select], [guide_md])
-    btn.click(train_model, [ml_ds, ml_feat], [plot_out, txt_out])
+    # --- EVENTS ---
+    # Explorer
+    ds_sel.change(update_explorer, [ds_sel, sp_sel], [sp_sel, data_view, c_raw, c_tr, meta_txt])
+    
+    # ML Tab: Обновление списка метрик при смене датасета
+    ml_ds_sel.change(sync_ml_metrics, [ml_ds_sel], [ml_feat_sel])
+    train_btn.click(train_model, [ml_ds_sel, ml_feat_sel], [p_out, t_out])
+    
+    # Methodology
+    meth_ds_sel.change(get_methodology_content, [meth_ds_sel], [meth_md])
     
-    # INITIAL LOAD
-    demo.load(update_explorer, [ds_select, split_select], [split_select, data_table, code_raw, code_tr, metadata_info])
-    demo.load(get_methodology_content, [method_ds_select], [guide_md])
+    # Initial Load
+    demo.load(update_explorer, [ds_sel, sp_sel], [sp_sel, data_view, c_raw, c_tr, meta_txt])
+    demo.load(sync_ml_metrics, [ml_ds_sel], [ml_feat_sel])
+    demo.load(get_methodology_content, [meth_ds_sel], [meth_md])
 
 if __name__ == "__main__":
     demo.launch()