Pareil

src/config/GPT2_Fine_Tuning_model_report_gpt_2_simple.json

@@ -0,0 +1,30 @@
+{
+  "created_at": "2025-12-07T16:23:15.572017Z",
+  "task": "language_modeling",
+  "target": "next_token_prediction",
+  "framework": "gpt_2_simple",
+  "dataset": {
+    "file": "\\fine_tuning\\description.txt",
+    "size_bytes": 769167,
+    "tokens": 208095
+  },
+  "model": {
+    "base_model_name": "124M",
+    "run_name": "run3",
+    "checkpoint_dir": "\\checkpoint\\run3",
+    "total_params": 0,
+    "size_mb": 0.0
+  },
+  "training": {
+    "steps": 200,
+    "restore_from": "fresh",
+    "batch_size": 1,
+    "learning_rate": 0.0001,
+    "optimizer": "Adam"
+  },
+  "metrics": {
+    "loss": 2.04,
+    "avg": 2.29,
+    "notes": "No explicit loss tracking from gpt_2_simple; metrics left to None."
+  }
+}

src/config/execution_generator_model_report.json

@@ -0,0 +1,31 @@
+{
+  "created_at": "2025-12-07T17:13:48.852853Z",
+  "task": "language_modeling",
+  "target": "next_token_prediction (execution_generation)",
+  "n_train_samples": 1347,
+  "n_val_samples": 150,
+  "sequence_length": 128,
+  "device": "cpu",
+  "model": {
+    "name": "transformer_execution_generator_v3",
+    "type": "CausalLM_FineTuned",
+    "class": "GPT2LMHeadModel",
+    "checkpoint_base": "gpt2",
+    "total_params": 124439808,
+    "trainable_params": 124439808,
+    "size_mb": 474.7
+  },
+  "training": {
+    "epochs": 8,
+    "learning_rate": 3e-05,
+    "batch_size": 32,
+    "seed_global": 42,
+    "optimizer": "AdamW"
+  },
+  "metrics": {
+    "final_train_loss": 1.2690867379654285,
+    "final_val_loss": 1.293086338043213,
+    "best_val_loss": 1.293086338043213,
+    "best_loss": 1.293086338043213
+  }
+}

src/config/model_report.json

@@ -0,0 +1,58 @@
+{
+  "created_at": "2025-12-03T09:04:03.083507Z",
+  "target": "Experience_Level",
+  "n_features": 8,
+  "n_test_samples": 2999,
+  "metrics_by_model": {
+    "Linear Regression": {
+      "MAE": 1.0702313301015012,
+      "RMSE": 1.3044658413432206,
+      "R2": -0.7023841026300084
+    },
+    "Random Forest": {
+      "MAE": 0.037401182031597605,
+      "RMSE": 0.09026074714124309,
+      "R2": 0.9918493924787138
+    },
+    "Bagging Regressor": {
+      "MAE": 0.03729884069670664,
+      "RMSE": 0.09030304487293238,
+      "R2": 0.9918417516603214
+    },
+    "Gradient Boosting": {
+      "MAE": 0.330239336587384,
+      "RMSE": 0.42489344861201594,
+      "R2": 0.8193856708756857
+    },
+    "KNN Regressor": {
+      "MAE": 0.15201366322234064,
+      "RMSE": 0.31524115755224025,
+      "R2": 0.9005790382902003
+    }
+  },
+  "selected_model": {
+    "model_type": "Random Forest",
+    "model_class": "RandomForestRegressor",
+    "model_path": "C:\\Users\\fback\\Desktop\\Projets\\Dev\\GitHub\\train.me\\src\\models\\v1\\life_style_data\\model.joblib",
+    "params": {
+      "bootstrap": true,
+      "ccp_alpha": 0.0,
+      "criterion": "squared_error",
+      "max_depth": null,
+      "max_features": 1.0,
+      "max_leaf_nodes": null,
+      "max_samples": null,
+      "min_impurity_decrease": 0.0,
+      "min_samples_leaf": 1,
+      "min_samples_split": 2,
+      "min_weight_fraction_leaf": 0.0,
+      "monotonic_cst": null,
+      "n_estimators": 200,
+      "n_jobs": null,
+      "oob_score": false,
+      "random_state": 42,
+      "verbose": 0,
+      "warm_start": false
+    }
+  }
+}

src/gradio/config.py

@@ -5,7 +5,7 @@ from huggingface_hub import hf_hub_download
 
 # ---- Chaînes centralisées (sans logique) ----
 # Dossiers / fichiers "métier"
-MODEL_SUBDIR = ("models", "v1", "life_style_data")
+MODEL_SUBDIR = ("config")
 # MODEL_FILENAME = "model.joblib"
 # SCHEMA_FILENAME = "feature_schema.json"
 REPORT_FILENAME = "model_report.json"

src/gradio/generators/execution_generator.py

@@ -13,7 +13,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 # ---------------------------------------------------------------------
 
 PROJECT_ROOT = Path(__file__).resolve().parents[2]
-MODEL_DIR = PROJECT_ROOT / "models" / "v1"
+MODEL_DIR = PROJECT_ROOT / "config"
 
 # Dossier de ton modèle finetuné d'exécution
 # EXEC_MODEL_DIR = MODEL_DIR / "transformer_execution_generator_v3"

src/gradio/pages/ml_tab.py

@@ -111,14 +111,24 @@ def render_ml_tab(
                     precision=2,
                 )
 
-                # Nouveau champ texte interprétation du niveau
+                # # Nouveau champ texte interprétation du niveau
+                # level_out = gr.Textbox(
+                #     label="Physical level (text)",
+                #     interactive=False,
+                #     lines=1,
+                #     max_lines=1,
+                # )
+
                 level_out = gr.Textbox(
                     label="Physical level (text)",
-                    interactive=False,
+                    value="",  # ou "—" si tu veux un placeholder visuel
+                    placeholder="Prediction will appear here",
+                    interactive=True,
                     lines=1,
                     max_lines=1,
                 )
 
+
                 meta_out = gr.Textbox(
                     label="Informations",
                     interactive=False,
@@ -153,6 +163,22 @@ def render_ml_tab(
         gr.Markdown("---")
 
         # ====== Prédiction ======
+        def _to_scalar(x):
+            """
+            Transforme n'importe quoi (np.array, liste, scalaire) en float Python.
+            Retourne None si on ne peut vraiment rien faire.
+            """
+            try:
+                if isinstance(x, np.ndarray):
+                    if x.shape == ():
+                        return float(x)
+                    if x.size == 1:
+                        return float(x.ravel()[0])
+                if isinstance(x, (list, tuple)) and len(x) == 1:
+                    return float(x[0])
+                return float(x)
+            except Exception:
+                return None
 
         def _interpret_level(y_val) -> str:
             """Map numeric prediction to textual level."""
@@ -203,7 +229,7 @@ def render_ml_tab(
                     payload["Workout_Frequency (days/week)"]
                 )
 
-            y_xp, meta = predict_single(
+            y_xp_raw, meta = predict_single(
                 payload=payload,
                 internal_expected=internal_expected,
                 model=model,
@@ -216,16 +242,23 @@ def render_ml_tab(
                 encoder=encoder,
             )
 
-            print("[DEBUG] y_xp =", y_xp)
+            # --- sécuriser la valeur numérique pour Gradio ---
+            y_xp = _to_scalar(y_xp_raw)
+            if y_xp is None:
+                y_xp_display = 0.0
+            else:
+                y_xp_display = round(float(y_xp), 2)
+
+            print("[DEBUG] y_xp_raw =", y_xp_raw)
+            print("[DEBUG] y_xp_scalar =", y_xp_display)
             print("[DEBUG] meta =", meta)
 
-            level_text = xp_to_label_safe(y_xp)
+            level_text = xp_to_label_safe(y_xp_display)
             info_text = str(meta)
 
             print("[DEBUG] level_text =", level_text)
             print("[DEBUG] info_text  =", info_text)
 
-
             # 2) Calcul BMI & Body Fat %
             bmi = None
             fat_pct = None
@@ -240,9 +273,7 @@ def render_ml_tab(
 
                 if bmi is not None and age_val not in (None, ""):
                     age_f = float(age_val)
-                    # 1 = homme, 0 = femme (IMG formule classique)
                     sex_flag = 1.0 if gender_raw.startswith("m") else 0.0
-
                     fat_pct = 1.20 * bmi + 0.23 * age_f - 10.8 * sex_flag - 5.4
                     fat_pct = round(fat_pct, 2)
             except Exception:
@@ -250,7 +281,7 @@ def render_ml_tab(
                 fat_pct = None
 
             # 4) Retourner les 5 sorties Gradio
-            return y_xp, level_text, bmi, fat_pct, meta
+            return y_xp_display, level_text, bmi, fat_pct, info_text
 
         btn.click(_fn, comps, [y_out, level_out, bmi_out, fat_out, meta_out])