script for loading music chunks as numpy arrays into tensorflow. fixes …

…#9

load_data.py

@@ -0,0 +1,70 @@
+from __future__ import absolute_import, division, print_function, unicode_literals
+
+import numpy as np
+import os
+import tensorflow as tf
+
+
+OUTPUT_DIR = "data/processed"
+
+filenames = [f
+             for f
+             in os.listdir(OUTPUT_DIR)
+             if os.path.isfile(os.path.join(OUTPUT_DIR, f))]
+
+## Getting the labels
+labels = [f.split("---")[0] for f in filenames]
+unique_labels = list(set(labels))
+num_labels = np.array([unique_labels.index(label) for label in labels])
+
+
+## Getting and stacking the data
+data = np.stack([np.load(os.path.join(OUTPUT_DIR,f),
+                         allow_pickle=True)
+                 for f
+                 in filenames],
+                0)
+
+
+p = np.random.permutation(len(num_labels))
+print(len(num_labels))
+print(len(p))
+num_labels = num_labels[p]
+data = data[p,:,:]
+
+train_count = int(len(labels) * 2 / 3)
+
+train_examples = data[:train_count]
+train_labels = num_labels[:train_count]
+test_examples = data[:train_count]
+test_labels = num_labels[:train_count]
+
+
+train_dataset = tf.data.Dataset.from_tensor_slices((train_examples, train_labels))
+test_dataset = tf.data.Dataset.from_tensor_slices((test_examples, test_labels))
+
+#train_examples = []
+#train_labels = []
+#test_examples = []
+#test_labels = []
+
+BATCH_SIZE = 64
+SHUFFLE_BUFFER_SIZE = 100
+
+train_dataset = train_dataset.shuffle(SHUFFLE_BUFFER_SIZE).batch(BATCH_SIZE)
+test_dataset = test_dataset.batch(BATCH_SIZE)
+
+
+model = tf.keras.Sequential([
+    tf.keras.layers.Flatten(input_shape=(1025, 44)),
+    tf.keras.layers.Dense(1000, activation='relu'),
+    tf.keras.layers.Dense(50, activation='relu'),
+    tf.keras.layers.Dense(3, activation='softmax')
+])
+
+model.compile(optimizer=tf.keras.optimizers.RMSprop(),
+                loss=tf.keras.losses.SparseCategoricalCrossentropy(),
+                metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])
+
+
+model.evaluate(test_dataset)

preprocess.py

@@ -37,7 +37,7 @@ def chunks(l, n):
     for c in chunks(song,segment_duration):
         if len(c) < segment_duration:
             continue
-        
+
         print(f"Writing DB array for {songfile} chunk {i}")
         chunk_path_out = os.path.join(OUTPUT_DIR,
                                       f"{songfile[:-4]}---{i}")