genarate accuracy & loss graph

backend/IT18257632/prepare_dataset.py

-import json
-import librosa
-import os
-
-DATASET_PATH = "dataset"
-JSON_PATH = "data.json"
-SAMPLES_TO_CONSIDER = 22050
-
-
-def prepare_dataset(dataset_path, json_path, n_mfcc=13, hop_length=512, n_fft=2048):
-    # data dictionary
-    data = {
-        "mappings": [],
-        "labels": [],
-        "MFCCs": [],
-        "files": []
-    }
-
-    # loop through all sub-dirs
-    for i, (dirpath, dirnames, filenames) in enumerate(os.walk(dataset_path)):
-
-        if dirpath is not dataset_path:
-
-            # update mapping
-            category = dirpath.split("/")
-            data["mappings"].append(category)
-            print(f"Processing{category}")
-
-            # loop through aii the file name and extract MFCCs
-            for f in filenames:
-
-                # get file path
-                file_path = os.path.join(dirpath, f)
-
-                # load audio file
-                signal, sr = librosa.load(file_path)
-
-                # ensure the audio file is at least 1 sec
-                if len(signal) >= SAMPLES_TO_CONSIDER:
-                    # enforce 1 sec signal
-                    signal = signal[:SAMPLES_TO_CONSIDER]
-
-                    # extract the MFCCs
-                    MFCCs = librosa.feature.mfcc(signal, sr, n_mfcc=n_mfcc, hop_length=hop_length, n_fft=n_fft)
-
-                    # store data
-                    data["labels"].append(i - 1)
-                    data["MFCCs"].append(MFCCs.T.tolist())
-                    data["files"].append(file_path)
-
-    with open(json_path, "w") as fp:
-        json.dump(data, fp, indent=4)
-
-
-if __name__ == "__main__":
-    prepare_dataset(DATASET_PATH, JSON_PATH)
+import json
+import librosa
+import os
+
+DATASET_PATH = "dataset"
+JSON_PATH = "data.json"
+SAMPLES_TO_CONSIDER = 22050
+
+
+def prepare_dataset(dataset_path, json_path, n_mfcc=13, hop_length=512, n_fft=2048):
+    # data dictionary
+    data = {
+        "mappings": [],
+        "labels": [],
+        "MFCCs": [],
+        "files": []
+    }
+
+    # loop through all sub-dirs
+    for i, (dirpath, dirnames, filenames) in enumerate(os.walk(dataset_path)):
+
+        if dirpath is not dataset_path:
+
+            # update mapping
+            category = dirpath.split("/")[-1]
+            data["mappings"].append(category)
+            print(f"Processing{category}")
+
+            # loop through aii the file name and extract MFCCs
+            for f in filenames:
+
+                # get file path
+                file_path = os.path.join(dirpath, f)
+
+                # load audio file
+                signal, sr = librosa.load(file_path)
+
+                # ensure the audio file is at least 1 sec
+                if len(signal) >= SAMPLES_TO_CONSIDER:
+                    # enforce 1 sec signal
+                    signal = signal[:SAMPLES_TO_CONSIDER]
+
+                    # extract the MFCCs
+                    MFCCs = librosa.feature.mfcc(signal, n_mfcc=n_mfcc, hop_length=hop_length, n_fft=n_fft)
+
+                    # store data
+                    data["labels"].append(i - 1)
+                    data["MFCCs"].append(MFCCs.T.tolist())
+                    data["files"].append(file_path)
+                    print(f"{file_path}: {i-1}")
+
+    # store in json file
+    with open(json_path, "w") as fp:
+        json.dump(data, fp, indent=4)
+
+
+if __name__ == "__main__":
+    prepare_dataset(DATASET_PATH, JSON_PATH)

backend/IT18257632/train.py → backend/IT18257632/train_model.py

 import json
 import numpy as np
+import matplotlib.pyplot as pyplot
 import tensorflow.keras as keras
 from sklearn.model_selection import train_test_split
 
@@ -7,10 +8,10 @@ DATA_PATH = "data.json"
 SAVE_MODEL_PATH = "model.h5"
 
 LEARNING_RATE = 0.0001
-EPOCHS = 40
+EPOCHS = 100
 BATCH_SIZE = 32
 
-NUM_KEYWORDS = 15
+NUM_KEYWORDS = 13
 
 
 def load_dataset(data_path):
@@ -78,6 +79,23 @@ def build_model(input_shape, learning_rate, error="sparse_categorical_crossentro
     return model
 
 
+def generate_graph(history):
+    pyplot.plot(history.history['loss'], 'r', label='train loss')
+    pyplot.plot(history.history['val_loss'], 'g', label='validation loss')
+    pyplot.legend()
+    pyplot.xlabel('Epochs')
+    pyplot.ylabel('Loss')
+    pyplot.show()
+
+    pyplot.subplot(1, 2, 1)
+    pyplot.ylabel('Accuracy')
+    pyplot.xlabel('Epochs')
+    pyplot.plot(history.history['accuracy'], label='Training Accuracy')
+    pyplot.plot(history.history['val_accuracy'], label='Validation Accuracy')
+    pyplot.legend()
+    pyplot.show()
+
+
 def main():
     # load train/validation/test data splits
     X_train, X_validation, X_test, y_train, y_validation, y_test = get_data_splits(DATA_PATH)
@@ -87,7 +105,8 @@ def main():
     model = build_model(input_shape, LEARNING_RATE)
 
     # train the model
-    model.fit(X_train, y_train, epochs=EPOCHS, batch_size=BATCH_SIZE, validation_data=(X_validation, y_validation))
+    history = model.fit(X_train, y_train, epochs=EPOCHS, batch_size=BATCH_SIZE,
+                        validation_data=(X_validation, y_validation))
 
     # evaluate the model
     test_error, test_accuracy = model.evaluate(X_test, y_test)
@@ -96,6 +115,8 @@ def main():
     # save the model
     model.save(SAVE_MODEL_PATH)
 
+    generate_graph(history)
+
 
 if __name__ == "__main__":
     main()