added model comparaison function

app.py

@@ -3,6 +3,7 @@ import pandas as pd
 from transformers import AutoModelForSequenceClassification, AutoTokenizer
 from datasets import load_dataset
 from sklearn.metrics import accuracy_score, precision_recall_fscore_support, confusion_matrix
+from sklearn.cluster import KMeans
 import torch
 from sentence_transformers import SentenceTransformer
 import umap
@@ -88,6 +89,54 @@ def generate_embeddings_and_plot(categories):
     tsne_plot_path = plot_embeddings(tsne_embeddings, "t-SNE Projection of Text Categories", "tsne")
     return umap_plot_path, tsne_plot_path
 
+def compare_models(model1, model2):
+    # Assuming dataset['test']['text'] returns a list of strings:
+    test_texts = dataset['test']['text']  # This is directly usable if it's a list
+    # Directly use the labels as a list, without calling .tolist()
+    labels = dataset['test']['label']
+
+    inputs1 = encode(test_texts, tokenizers[model1])
+    inputs2 = encode(test_texts, tokenizers[model2])
+
+    preds1 = predict(models[model1], inputs1)
+    preds2 = predict(models[model2], inputs2)
+
+    metrics1 = calculate_metrics(labels, preds1)
+    metrics2 = calculate_metrics(labels, preds2)
+
+    categories = {
+        "correct_both": [],
+        "incorrect_both": [],
+        "correct_model1_only": [],
+        "correct_model2_only": []
+    }
+
+    for i, label in enumerate(labels):
+        text = test_texts[i]
+        if preds1[i] == label and preds2[i] == label:
+            categories["correct_both"].append(text)
+        elif preds1[i] != label and preds2[i] != label:
+            categories["incorrect_both"].append(text)
+        elif preds1[i] == label and preds2[i] != label:
+            categories["correct_model1_only"].append(text)
+        elif preds1[i] != label and preds2[i] == label:
+            categories["correct_model2_only"].append(text)
+
+    # Generate metrics DataFrame
+    metrics_df = pd.DataFrame({
+        "Metric": ["Accuracy", "Precision", "Recall", "F1 Score"],
+        model1: metrics1[:-1],
+        model2: metrics2[:-1],
+    })
+    metrics_df["% Difference"] = ((metrics_df[model1] - metrics_df[model2]) / metrics_df[model2] * 100).apply(lambda x: f"{x:.2f}%")
+    
+    # Confusion matrices and visualizations
+    conf_matrix_path1 = generate_confusion_matrix(metrics1[-1], model1)
+    conf_matrix_path2 = generate_confusion_matrix(metrics2[-1], model2)
+    umap_plot_path, tsne_plot_path = generate_embeddings_and_plot(categories)
+
+    return metrics_df, conf_matrix_path1, conf_matrix_path2, umap_plot_path, tsne_plot_path, categories
+
 def setup_gradio_interface():
     with gr.Blocks() as demo:
         gr.Markdown("## Model Comparison and Text Analysis")