Update app.py

app.py

@@ -7,178 +7,6 @@ import seaborn as sns
 import io
 import gradio as gr
 import pickle
-# Load fine-tuned results from pickle
-finetuned_pickle_path = 'best_params_backup.pkl'
-
-with open(finetuned_pickle_path, 'rb') as f:
-    best_params_per_dataset = pickle.load(f)
-
-# Convert to DataFrame with accuracy only
-ft_rows = []
-for key, val in best_params_per_dataset.items():
-    parts = key.split("_", 1)
-    dataset = parts[0]
-    model = parts[1] if len(parts) > 1 else "unknown"
-    ft_rows.append({
-        "dataset": dataset,
-        "model": model,
-        "accuracy": round(val["score"], 4)
-    })
-
-finetuned_df = pd.DataFrame(ft_rows)
-
-# Build fine-tuned pairwise comparisons — all datasets
-ft_models = ["Random Forest", "Decision Tree", "SVM", "KNN", "NN"]
-
-ft_all_results = []
-ft_model_list = ft_models.copy()
-
-for model_a in ft_models:
-    other = ft_model_list.copy()
-    other.remove(model_a)
-    for model_b in other:
-        data_a = finetuned_df[finetuned_df['model'] == model_a].set_index('dataset')['accuracy']
-        data_b = finetuned_df[finetuned_df['model'] == model_b].set_index('dataset')['accuracy']
-        combined = pd.DataFrame({'a': data_a, 'b': data_b}).dropna()
-        if len(combined) < 2:
-            continue
-        t_stat, p_val = ttest_rel(combined['a'], combined['b'])
-        ft_all_results.append({
-            'model_a': model_a,
-            'model_b': model_b,
-            'mean_a': combined['a'].mean(),
-            'mean_b': combined['b'].mean(),
-            'std_a': combined['a'].std(),
-            'std_b': combined['b'].std(),
-            'n_datasets': len(combined),
-            't_statistic': t_stat,
-            'p_value': p_val
-        })
-    ft_model_list = other.copy()
-
-ft_results_df = pd.DataFrame(ft_all_results)
-
-# Build per-category fine-tuned comparisons
-ft_sig = {}
-for cat_key in list(DATASET_CATEGORIES.keys()):
-    cat_datasets = list(DATASET_CATEGORIES[cat_key].keys())
-    df_cat = finetuned_df[finetuned_df['dataset'].isin(cat_datasets)]
-    cat_results = []
-    ft_model_list = ft_models.copy()
-    for model_a in ft_models:
-        other = ft_model_list.copy()
-        other.remove(model_a)
-        for model_b in other:
-            data_a = df_cat[df_cat['model'] == model_a].set_index('dataset')['accuracy']
-            data_b = df_cat[df_cat['model'] == model_b].set_index('dataset')['accuracy']
-            combined = pd.DataFrame({'a': data_a, 'b': data_b}).dropna()
-            if len(combined) < 2:
-                continue
-            t_stat, p_val = ttest_rel(combined['a'], combined['b'])
-            cat_results.append({
-                'model_a': model_a,
-                'model_b': model_b,
-                'mean_a': combined['a'].mean(),
-                'mean_b': combined['b'].mean(),
-                'std_a': combined['a'].std(),
-                'std_b': combined['b'].std(),
-                'n_datasets': len(combined),
-                't_statistic': t_stat,
-                'p_value': p_val
-            })
-        ft_model_list = other.copy()
-    ft_sig[cat_key] = pd.DataFrame(cat_results)
-
-ft_sig["AllDatasets"] = ft_results_df
-def get_keys(d, values):
-  return [k for k, v in d.items() if v in values]
-pickle_file_path = 'model_results1.pkl'
-
-model_results = pd.read_pickle(pickle_file_path)
-
-csv_file_path = 'the_model_results.csv'
-
-model_results_csv = pd.read_csv(csv_file_path)
-
-fmodel_results = pd.concat([model_results, model_results_csv.rename(columns = {"dataset_name" : "dataset"})], ignore_index=True)
-
-def ft_compare_groups(data_choice, model1, model2):
-    data1 = ft_sig[data_choice.split(' (')[0]]
-    comparison_data = data1[
-        ((data1['model_a'] == model1) & (data1['model_b'] == model2)) |
-        ((data1['model_a'] == model2) & (data1['model_b'] == model1))
-    ]
-
-    if comparison_data.empty:
-        fig = plt.figure(figsize=(10, 6))
-        plt.close(fig)
-        return fig, "No comparison data found. Don't pick the same models."
-
-    plot_data = []
-    p_values_text = []
-
-    for _, row in comparison_data.iterrows():
-        if row['model_a'] == model1:
-            plot_data.append({'Model': model1, 'Mean Accuracy': row['mean_a']})
-            plot_data.append({'Model': model2, 'Mean Accuracy': row['mean_b']})
-        else:
-            plot_data.append({'Model': model1, 'Mean Accuracy': row['mean_b']})
-            plot_data.append({'Model': model2, 'Mean Accuracy': row['mean_a']})
-        p_values_text.append(
-            f"accuracy p-value: {row['p_value']:.5f} "
-            f"(Significant (cutoff = 0.05): {'Yes' if row['p_value'] < 0.05 else 'No'})"
-        )
-
-    df_plot = pd.DataFrame(plot_data)
-
-    fig = plt.figure(figsize=(10, 6))
-    sns.barplot(x='Model', y='Mean Accuracy', data=df_plot)
-    plt.title(f'Fine-Tuned: {model1} vs {model2} — Accuracy')
-    plt.ylabel('Mean Accuracy')
-    plt.xlabel('Model')
-    plt.ylim(0, 1)
-    plt.tight_layout()
-
-    return fig, "\n".join(p_values_text)
-
-
-def ft_compare_ind(med, game, ed, bank, sci, social, ml, other, models_to_compare=None):
-    selected_keys = []
-    dropdowns = [med, game, ed, bank, sci, social, ml, other]
-
-    for cat_name, dropdown_values in zip(cats1, dropdowns):
-        if dropdown_values:
-            selected_keys.extend(get_keys(DATASET_CATEGORIES[cat_name], dropdown_values))
-
-    if not models_to_compare:
-        models_to_compare = ft_models
-
-    dataset_id_to_name = {
-        id: name
-        for category_dict in DATASET_CATEGORIES.values()
-        for id, name in category_dict.items()
-    }
-
-    filtered_df = finetuned_df[
-        (finetuned_df["dataset"].isin(selected_keys)) &
-        (finetuned_df["model"].isin(models_to_compare))
-    ].copy()
-
-    heatmap_data = filtered_df.pivot_table(
-        index='dataset',
-        columns='model',
-        values='accuracy'
-    )
-    heatmap_data = heatmap_data.rename(index=dataset_id_to_name)
-
-    fig = plt.figure(figsize=(12, 8))
-    sns.heatmap(heatmap_data, annot=True, cmap="crest", fmt=".3f", cbar=True)
-    plt.title(f"Fine-Tuned Accuracy per Dataset and Model ({len(selected_keys)} datasets)")
-    plt.ylabel("Dataset")
-    plt.xlabel("Model")
-    plt.tight_layout()
-
-    return fig, "Comparison complete."
 DATASET_CATEGORIES = {
     "Medical & Healthcare": {
         "D1": "Heart Disease (Comprehensive)",
@@ -401,6 +229,179 @@ DATASET_CATEGORIES = {
     }
 }
 cats1 = list(DATASET_CATEGORIES.keys())
+# Load fine-tuned results from pickle
+finetuned_pickle_path = 'best_params_backup.pkl'
+
+with open(finetuned_pickle_path, 'rb') as f:
+    best_params_per_dataset = pickle.load(f)
+
+# Convert to DataFrame with accuracy only
+ft_rows = []
+for key, val in best_params_per_dataset.items():
+    parts = key.split("_", 1)
+    dataset = parts[0]
+    model = parts[1] if len(parts) > 1 else "unknown"
+    ft_rows.append({
+        "dataset": dataset,
+        "model": model,
+        "accuracy": round(val["score"], 4)
+    })
+
+finetuned_df = pd.DataFrame(ft_rows)
+
+# Build fine-tuned pairwise comparisons — all datasets
+ft_models = ["Random Forest", "Decision Tree", "SVM", "KNN", "NN"]
+
+ft_all_results = []
+ft_model_list = ft_models.copy()
+
+for model_a in ft_models:
+    other = ft_model_list.copy()
+    other.remove(model_a)
+    for model_b in other:
+        data_a = finetuned_df[finetuned_df['model'] == model_a].set_index('dataset')['accuracy']
+        data_b = finetuned_df[finetuned_df['model'] == model_b].set_index('dataset')['accuracy']
+        combined = pd.DataFrame({'a': data_a, 'b': data_b}).dropna()
+        if len(combined) < 2:
+            continue
+        t_stat, p_val = ttest_rel(combined['a'], combined['b'])
+        ft_all_results.append({
+            'model_a': model_a,
+            'model_b': model_b,
+            'mean_a': combined['a'].mean(),
+            'mean_b': combined['b'].mean(),
+            'std_a': combined['a'].std(),
+            'std_b': combined['b'].std(),
+            'n_datasets': len(combined),
+            't_statistic': t_stat,
+            'p_value': p_val
+        })
+    ft_model_list = other.copy()
+
+ft_results_df = pd.DataFrame(ft_all_results)
+
+# Build per-category fine-tuned comparisons
+ft_sig = {}
+for cat_key in list(DATASET_CATEGORIES.keys()):
+    cat_datasets = list(DATASET_CATEGORIES[cat_key].keys())
+    df_cat = finetuned_df[finetuned_df['dataset'].isin(cat_datasets)]
+    cat_results = []
+    ft_model_list = ft_models.copy()
+    for model_a in ft_models:
+        other = ft_model_list.copy()
+        other.remove(model_a)
+        for model_b in other:
+            data_a = df_cat[df_cat['model'] == model_a].set_index('dataset')['accuracy']
+            data_b = df_cat[df_cat['model'] == model_b].set_index('dataset')['accuracy']
+            combined = pd.DataFrame({'a': data_a, 'b': data_b}).dropna()
+            if len(combined) < 2:
+                continue
+            t_stat, p_val = ttest_rel(combined['a'], combined['b'])
+            cat_results.append({
+                'model_a': model_a,
+                'model_b': model_b,
+                'mean_a': combined['a'].mean(),
+                'mean_b': combined['b'].mean(),
+                'std_a': combined['a'].std(),
+                'std_b': combined['b'].std(),
+                'n_datasets': len(combined),
+                't_statistic': t_stat,
+                'p_value': p_val
+            })
+        ft_model_list = other.copy()
+    ft_sig[cat_key] = pd.DataFrame(cat_results)
+
+ft_sig["AllDatasets"] = ft_results_df
+def get_keys(d, values):
+  return [k for k, v in d.items() if v in values]
+pickle_file_path = 'model_results1.pkl'
+
+model_results = pd.read_pickle(pickle_file_path)
+
+csv_file_path = 'the_model_results.csv'
+
+model_results_csv = pd.read_csv(csv_file_path)
+
+fmodel_results = pd.concat([model_results, model_results_csv.rename(columns = {"dataset_name" : "dataset"})], ignore_index=True)
+
+def ft_compare_groups(data_choice, model1, model2):
+    data1 = ft_sig[data_choice.split(' (')[0]]
+    comparison_data = data1[
+        ((data1['model_a'] == model1) & (data1['model_b'] == model2)) |
+        ((data1['model_a'] == model2) & (data1['model_b'] == model1))
+    ]
+
+    if comparison_data.empty:
+        fig = plt.figure(figsize=(10, 6))
+        plt.close(fig)
+        return fig, "No comparison data found. Don't pick the same models."
+
+    plot_data = []
+    p_values_text = []
+
+    for _, row in comparison_data.iterrows():
+        if row['model_a'] == model1:
+            plot_data.append({'Model': model1, 'Mean Accuracy': row['mean_a']})
+            plot_data.append({'Model': model2, 'Mean Accuracy': row['mean_b']})
+        else:
+            plot_data.append({'Model': model1, 'Mean Accuracy': row['mean_b']})
+            plot_data.append({'Model': model2, 'Mean Accuracy': row['mean_a']})
+        p_values_text.append(
+            f"accuracy p-value: {row['p_value']:.5f} "
+            f"(Significant (cutoff = 0.05): {'Yes' if row['p_value'] < 0.05 else 'No'})"
+        )
+
+    df_plot = pd.DataFrame(plot_data)
+
+    fig = plt.figure(figsize=(10, 6))
+    sns.barplot(x='Model', y='Mean Accuracy', data=df_plot)
+    plt.title(f'Fine-Tuned: {model1} vs {model2} — Accuracy')
+    plt.ylabel('Mean Accuracy')
+    plt.xlabel('Model')
+    plt.ylim(0, 1)
+    plt.tight_layout()
+
+    return fig, "\n".join(p_values_text)
+
+
+def ft_compare_ind(med, game, ed, bank, sci, social, ml, other, models_to_compare=None):
+    selected_keys = []
+    dropdowns = [med, game, ed, bank, sci, social, ml, other]
+
+    for cat_name, dropdown_values in zip(cats1, dropdowns):
+        if dropdown_values:
+            selected_keys.extend(get_keys(DATASET_CATEGORIES[cat_name], dropdown_values))
+
+    if not models_to_compare:
+        models_to_compare = ft_models
+
+    dataset_id_to_name = {
+        id: name
+        for category_dict in DATASET_CATEGORIES.values()
+        for id, name in category_dict.items()
+    }
+
+    filtered_df = finetuned_df[
+        (finetuned_df["dataset"].isin(selected_keys)) &
+        (finetuned_df["model"].isin(models_to_compare))
+    ].copy()
+
+    heatmap_data = filtered_df.pivot_table(
+        index='dataset',
+        columns='model',
+        values='accuracy'
+    )
+    heatmap_data = heatmap_data.rename(index=dataset_id_to_name)
+
+    fig = plt.figure(figsize=(12, 8))
+    sns.heatmap(heatmap_data, annot=True, cmap="crest", fmt=".3f", cbar=True)
+    plt.title(f"Fine-Tuned Accuracy per Dataset and Model ({len(selected_keys)} datasets)")
+    plt.ylabel("Dataset")
+    plt.xlabel("Model")
+    plt.tight_layout()
+
+    return fig, "Comparison complete."
+
 import pandas as pd
 from scipy.stats import ttest_rel