Update src/streamlit_app.py

src/streamlit_app.py

@@ -1,13 +1,15 @@
 import streamlit as st
 from pathlib import Path
 
-from app.inference import load_models, predict_image
+import matplotlib.pyplot as plt
+import numpy as np
+from sklearn.metrics import confusion_matrix
+
+from app.inference import load_models, predict_image, evaluate_dataset
 from app.config import DATA_ROOT, IMAGE_EXTENSIONS
 
 # Paths inside the Space
 PROJECT_ROOT = Path(__file__).resolve().parent
-REPORTS_DIR = PROJECT_ROOT / "reports"
-CONFUSION_IMAGE_PATH = REPORTS_DIR / "confusion_matrix_normalized.png"
 
 # -------------------------------------------------------------------
 # STATIC REPORT TEXTS (Block 8 and Block 10)
@@ -142,174 +144,7 @@ Predicting: 100%|█████████████████████
 📊 Overall Accuracy: 0.9975 (99.75%)
 📊 Average confidence: 0.8239
 ❌ Failed to process: 48 images
-================================================================================
-❌ WRONG PREDICTIONS DETAILS (44 total)
-================================================================================
-Classes with wrong predictions:
-            actual_class  wrong_count                                                        predicted_as
-        pins_Brie Larson            4            pins_Emma Stone, pins_ellen page, pins_Jennifer Lawrence
-     pins_Jessica Barden            3     pins_Alex Lawther, pins_kiernen shipka, pins_Danielle Panabaker
-       pins_Logan Lerman            3 pins_Sarah Wayne Callies, pins_Leonardo DiCaprio, pins_Eliza Taylor
- pins_scarlett johansson            2                                   pins_gal gadot, pins_Taylor Swift
-      pins_Emilia Clarke            2                           pins_Marie Avgeropoulos, pins_Irina Shayk
-   pins_Brenton Thwaites            2                           pins_Leonardo DiCaprio, pins_Bobby Morley
-    pins_elizabeth olsen            2                            pins_ellen page, pins_Millie Bobby Brown
-            pins_Zendaya            2                               pins_Lili Reinhart, pins_Selena Gomez
-        pins_Tom Holland            2                           pins_Anne Hathaway, pins_Robert Downey Jr
-     pins_Tom Hiddleston            1                                                    pins_Chris Evans
-       pins_Selena Gomez            1                                                    pins_Emma Watson
-       pins_Taylor Swift            1                                                     pins_Emma Stone
-        pins_Zoe Saldana            1                                                 pins_Lindsey Morgan
-     pins_Richard Harmon            1                                              pins_Leonardo DiCaprio
-        pins_amber heard            1                                                    pins_Brie Larson
-          pins_gal gadot            1                                               pins_Madelaine Petsch
-      pins_margot robbie            1                                                    pins_Emma Watson
-pins_Sarah Wayne Callies            1                                                 pins_Lindsey Morgan
-      pins_Avril Lavigne            1                                             pins_Alexandra Daddario
-    pins_Natalie Portman            1                                             pins_Millie Bobby Brown
-       pins_Nadia Hilker            1                                             pins_Marie Avgeropoulos
- pins_Marie Avgeropoulos            1                                                    pins_Johnny Depp
-     pins_Lindsey Morgan            1                                                  pins_Anne Hathaway
-  pins_Leonardo DiCaprio            1                                               pins_Brenton Thwaites
- pins_Katherine Langford            1                                               pins_Madelaine Petsch
-        pins_Johnny Depp            1                                              pins_Leonardo DiCaprio
-        pins_Irina Shayk            1                                                  pins_Maria Pedraza
-       pins_Hugh Jackman            1                                                      pins_tom ellis
-         pins_Emma Stone            1                                                  pins_margot robbie
-    pins_Chris Hemsworth            1                                                    pins_Chris Evans
-    pins_Morena Baccarin            1                                                  pins_camila mendes
---------------------------------------------------------------------------------
-Individual wrong predictions (showing first 20):
---------------------------------------------------------------------------------
-  • amber heard214_323.jpg
-    Actual: pins_amber heard → Predicted: pins_Brie Larson (confidence: 0.275)
-  • Avril Lavigne238_664.jpg
-    Actual: pins_Avril Lavigne → Predicted: pins_Alexandra Daddario (confidence: 0.131)
-  • Brenton Thwaites46_885.jpg
-    Actual: pins_Brenton Thwaites → Predicted: pins_Leonardo DiCaprio (confidence: 0.273)
-  • Brenton Thwaites99_936.jpg
-    Actual: pins_Brenton Thwaites → Predicted: pins_Bobby Morley (confidence: 0.204)
-  • Brie Larson157_994.jpg
-    Actual: pins_Brie Larson → Predicted: pins_Emma Stone (confidence: 0.136)
-  • Brie Larson172_1007.jpg
-    Actual: pins_Brie Larson → Predicted: pins_ellen page (confidence: 0.253)
-  • Brie Larson187_1021.jpg
-    Actual: pins_Brie Larson → Predicted: pins_Jennifer Lawrence (confidence: 0.127)
-  • Brie Larson77_1088.jpg
-    Actual: pins_Brie Larson → Predicted: pins_margot robbie (confidence: 0.330)
-  • Chris Hemsworth1_384.jpg
-    Actual: pins_Chris Hemsworth → Predicted: pins_Chris Evans (confidence: 0.095)
-  • elizabeth olsen164_1173.jpg
-    Actual: pins_elizabeth olsen → Predicted: pins_ellen page (confidence: 0.355)
-  • elizabeth olsen170_1179.jpg
-    Actual: pins_elizabeth olsen → Predicted: pins_Millie Bobby Brown (confidence: 0.272)
-  • Emilia Clarke194_952.jpg
-    Actual: pins_Emilia Clarke → Predicted: pins_Marie Avgeropoulos (confidence: 0.150)
-  • Emilia Clarke48_1021.jpg
-    Actual: pins_Emilia Clarke → Predicted: pins_Irina Shayk (confidence: 0.069)
-  • Emma Stone36_1779.jpg
-    Actual: pins_Emma Stone → Predicted: pins_margot robbie (confidence: 0.282)
-  • gal gadot134_1690.jpg
-    Actual: pins_gal gadot → Predicted: pins_Madelaine Petsch (confidence: 0.234)
-  • Hugh Jackman118_1288.jpg
-    Actual: pins_Hugh Jackman → Predicted: pins_tom ellis (confidence: 0.128)
-  • Irina Shayk236_2335.jpg
-    Actual: pins_Irina Shayk → Predicted: pins_Maria Pedraza (confidence: 0.082)
-  • Jessica Barden211_1449.jpg
-    Actual: pins_Jessica Barden → Predicted: pins_Alex Lawther (confidence: 0.779)
-  • Jessica Barden31_1475.jpg
-    Actual: pins_Jessica Barden → Predicted: pins_kiernen shipka (confidence: 0.098)
-  • Jessica Barden34_1478.jpg
-    Actual: pins_Jessica Barden → Predicted: pins_Danielle Panabaker (confidence: 0.048)
-  ... and 24 more wrong predictions (see CSV for details)
-================================================================================
-🎯 CLASSES WITH 100% ACCURACY (74 classes)
-================================================================================
-              class_name  total_count
-       pins_Adriana Lima          213
- pins_Millie Bobby Brown          191
-            pins_Rihanna          132
-   pins_Rebecca Ferguson          178
-         pins_Rami Malek          160
-       pins_Penn Badgley          171
-       pins_Pedro Alonso          125
-pins_Neil Patrick Harris          116
-     pins_Natalie Dormer          196
-     pins_Morgan Freeman          102
-        pins_Miley Cyrus          178
-       pins_Keanu Reeves          158
-          pins_Megan Fox          208
-    pins_Mark Zuckerberg           95
-       pins_Mark Ruffalo          177
-       pins_Alex Lawther          152
-    pins_Maisie Williams          193
-   pins_Madelaine Petsch          192
-       pins_Lionel Messi           86
-      pins_Lili Reinhart          150
-... and 54 more classes
-================================================================================
-⚠️ CLASSES WITH LOWEST ACCURACY (Bottom 10)
-================================================================================
-             class_name  correct_count  wrong_count  total_count  accuracy
-      pins_Taylor Swift            129            1          130  0.992308
-   pins_elizabeth olsen            219            2          221  0.990950
-  pins_Brenton Thwaites            207            2          209  0.990431
-     pins_Emilia Clarke            207            2          209  0.990431
-pins_scarlett johansson            199            2          201  0.990050
-       pins_Tom Holland            187            2          189  0.989418
-      pins_Logan Lerman            209            3          212  0.985849
-           pins_Zendaya            135            2          137  0.985401
-    pins_Jessica Barden            138            3          141  0.978723
-       pins_Brie Larson            165            4          169  0.976331
-================================================================================
-✅ CORRECT PREDICTIONS SAMPLE (showing 10 of 17442)
-================================================================================
-  ✓ Adriana Lima0_0.jpg: pins_Adriana Lima (confidence: 0.787)
-  ✓ Adriana Lima101_3.jpg: pins_Adriana Lima (confidence: 0.946)
-  ✓ Adriana Lima102_4.jpg: pins_Adriana Lima (confidence: 0.907)
-  ✓ Adriana Lima103_5.jpg: pins_Adriana Lima (confidence: 0.752)
-  ✓ Adriana Lima104_6.jpg: pins_Adriana Lima (confidence: 0.886)
-  ✓ Adriana Lima105_7.jpg: pins_Adriana Lima (confidence: 0.779)
-  ✓ Adriana Lima106_8.jpg: pins_Adriana Lima (confidence: 0.794)
-  ✓ Adriana Lima107_9.jpg: pins_Adriana Lima (confidence: 0.930)
-  ✓ Adriana Lima108_10.jpg: pins_Adriana Lima (confidence: 0.902)
-  ✓ Adriana Lima109_11.jpg: pins_Adriana Lima (confidence: 0.375)
-================================================================================
-📁 OUTPUT FILES SAVED:
-================================================================================
-✅ predictions_results.csv
-   → All predictions sorted (correct first, then wrong)
-   → Columns: filename, actual, predicted, confidence, status, top3
-✅ predictions_summary.csv
-   → Per-class accuracy summary
-   → Columns: class_name, correct_count, wrong_count, total_count, accuracy
-================================================================================
-================================================================================
-❌ FAILED TO PROCESS (48 images)
-================================================================================
-  • Anne Hathaway203_391.jpg: No face detected
-  • Avril Lavigne11_572.jpg: No face detected
-  • Avril Lavigne174_619.jpg: No face detected
-  • Avril Lavigne41_685.jpg: No face detected
-  • barbara palvin158_800.jpg: No face detected
-  • Cristiano Ronaldo209_1326.jpg: No face detected
-  • Cristiano Ronaldo226_1333.jpg: No face detected
-  • Eliza Taylor202_775.jpg: No face detected
-  • Elizabeth Lail102_1055.jpg: No face detected
-  • Elizabeth Lail102_1056.jpg: No face detected
-  • Elizabeth Lail194_1117.jpg: No face detected
-  • Emilia Clarke78_1050.jpg: No face detected
-  • Emma Stone73_1817.jpg: No face detected
-  • Hugh Jackman119_1289.jpg: No face detected
-  • jeff bezos112_2049.jpg: No face detected
-  • jeff bezos12_2052.jpg: No face detected
-  • jeff bezos160_2068.jpg: No face detected
-  • jeff bezos178_2073.jpg: No face detected
-  • Jeremy Renner175_2634.jpg: No face detected
-  • Johnny Depp23_1863.jpg: No face detected
-  ... and 28 more
-✅ Failed images list saved to: failed_predictions.csv
-================================================================================
+... (rest of your long text, keep as is)
 ✅ PROCESSING COMPLETE!
 ================================================================================
 """
@@ -466,7 +301,7 @@ if page == "Home":
         st.warning("Dataset not ready yet.")
         selected_image_path = None
     else:
-        # DATA_ROOT should be: .../face_recognition_dataset/face_recognition_dataset
+        # List person folders
         person_folders = sorted(
             [f for f in DATA_ROOT.iterdir() if f.is_dir()]
         )
@@ -480,7 +315,6 @@ if page == "Home":
                 "Select Person Folder", ["All"] + person_names
             )
 
-            # Collect images depending on selection
             if person_option == "All":
                 image_files = [
                     p
@@ -499,7 +333,6 @@ if page == "Home":
                 st.sidebar.warning("No images found in the selected folder.")
                 selected_image_path = None
             else:
-                # Show relative paths (class_name/image.jpg) to make dropdown clearer
                 display_names = [
                     img.relative_to(DATA_ROOT).as_posix() for img in image_files
                 ]
@@ -546,17 +379,33 @@ elif page == "Training Report":
     st.code(TRAINING_REPORT_TEXT, language="text")
 
     st.markdown("---")
-    st.subheader("Normalized Confusion Matrix")
+    st.subheader("Confusion Matrix (programmatically computed, subset)")
 
-    if CONFUSION_IMAGE_PATH.exists():
-        left, center, right = st.columns([1, 3, 1])
-        with center:
-            st.image(CONFUSION_IMAGE_PATH.as_posix(), use_container_width=True)
+    # Compute confusion matrix on a small subset to keep it lightweight
+    with st.spinner("Computing confusion matrix on a small subset of the dataset..."):
+        df, acc = evaluate_dataset(images_per_class=2, max_images=150)
+
+    if df is None or df.empty:
+        st.info("Could not compute confusion matrix: evaluation subset is empty.")
     else:
-        st.warning(
-            "Confusion matrix image not found. "
-            "Place `confusion_matrix_normalized.png` under `src/reports/`."
-        )
+        labels = sorted(df["true_label"].unique())
+        cm = confusion_matrix(df["true_label"], df["predicted_label"], labels=labels)
+
+        fig, ax = plt.subplots(figsize=(8, 8))
+        im = ax.imshow(cm, interpolation="nearest")
+        ax.set_xticks(np.arange(len(labels)))
+        ax.set_yticks(np.arange(len(labels)))
+        ax.set_xticklabels(labels, rotation=90, fontsize=5)
+        ax.set_yticklabels(labels, fontsize=5)
+        ax.set_xlabel("Predicted label")
+        ax.set_ylabel("True label")
+        ax.set_title("Confusion Matrix (subset)")
+        plt.tight_layout()
+
+        st.pyplot(fig)
+
+        if acc is not None:
+            st.caption(f"Subset accuracy on this evaluation run: {acc:.4f}")
 
 # -------------------------------------------------------------------
 # PREDICTION REPORT PAGE