File size: 5,193 Bytes
564354e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pickle, glob\n",
    "import numpy as np\n",
    "import keras.utils as keras_utils\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from keras.layers import Dropout\n",
    "from keras.layers import LSTM\n",
    "from keras.layers import BatchNormalization as BatchNorm\n",
    "from keras.layers import Activation\n",
    "from keras.callbacks import ModelCheckpoint, EarlyStopping\n",
    "from music21 import converter, instrument, note, chord"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# To generate a new list of notes to train against\n",
    "# notes = []\n",
    "# for file in glob.glob(\"./midi_songs/*.mid\"):\n",
    "#     midi = converter.parse(file)\n",
    "#     notes_to_parse = None\n",
    "#     parts = instrument.partitionByInstrument(midi)\n",
    "#     if parts: # file has instrument parts\n",
    "#         notes_to_parse = parts.parts[0].recurse()\n",
    "#     else: # file has notes in a flat structure\n",
    "#         notes_to_parse = midi.flat.notes\n",
    "#     for element in notes_to_parse:\n",
    "#         if isinstance(element, note.Note):\n",
    "#             notes.append(str(element.pitch))\n",
    "#         elif isinstance(element, chord.Chord):\n",
    "#             notes.append('.'.join(str(n) for n in element.normalOrder))\n",
    "\n",
    "# with open('data/music_notes.pkl', 'wb') as filepath:\n",
    "#     pickle.dump(notes, filepath)\n",
    "#     pickle.dump(pitchnames, filepath)\n",
    "#     pickle.dump(n_vocab, filepath)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open('data/music_notes.pkl', 'rb') as filepath:\n",
    "    notes = pickle.load(filepath)\n",
    "    pitchnames = pickle.load(filepath)\n",
    "    n_vocab = pickle.load(filepath)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "sequence_length = 100\n",
    "# get all pitch names\n",
    "pitchnames = sorted(set(item for item in notes))\n",
    "# create a dictionary to map pitches to integers\n",
    "note_to_int = dict((note, number) for number, note in enumerate(pitchnames))\n",
    "network_input = []\n",
    "network_output = []\n",
    "# create input sequences and the corresponding outputs\n",
    "for i in range(0, len(notes) - sequence_length, 1):\n",
    "    sequence_in = notes[i:i + sequence_length]\n",
    "    sequence_out = notes[i + sequence_length]\n",
    "    network_input.append([note_to_int[char] for char in sequence_in])\n",
    "    network_output.append(note_to_int[sequence_out])\n",
    "n_patterns = len(network_input)\n",
    "# reshape the input into a format compatible with LSTM layers\n",
    "network_input = np.reshape(network_input, (n_patterns, sequence_length, 1))\n",
    "# normalize input\n",
    "network_input = network_input / float(n_vocab)\n",
    "network_output = keras_utils.to_categorical(network_output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Sequential()\n",
    "model.add(LSTM(\n",
    "    512,\n",
    "    input_shape=(network_input.shape[1], network_input.shape[2]),\n",
    "    recurrent_dropout=0.3,\n",
    "    return_sequences=True\n",
    "))\n",
    "model.add(LSTM(512, return_sequences=True, recurrent_dropout=0.3,))\n",
    "model.add(LSTM(512))\n",
    "model.add(BatchNorm())\n",
    "model.add(Dropout(0.3))\n",
    "model.add(Dense(256))\n",
    "model.add(Activation('relu'))\n",
    "model.add(BatchNorm())\n",
    "model.add(Dropout(0.3))\n",
    "model.add(Dense(n_vocab))\n",
    "model.add(Activation('softmax'))\n",
    "model.compile(loss='categorical_crossentropy', optimizer='rmsprop')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "checkpoint_filepath = \"./models/model.keras\"    \n",
    "checkpoint = ModelCheckpoint(\n",
    "    filepath=checkpoint_filepath,\n",
    "    monitor='loss',\n",
    "    mode='min',\n",
    "    save_best_only=True,\n",
    "    verbose=0\n",
    ")\n",
    "early_stopping = EarlyStopping(\n",
    "    monitor=\"loss\",\n",
    "    patience=10,\n",
    "    min_delta=0.001,\n",
    "    restore_best_weights=True,\n",
    ")\n",
    "callbacks_list = [early_stopping, checkpoint]     \n",
    "model.fit(network_input, network_output, epochs=1000, batch_size=64, callbacks=callbacks_list)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": ".venv",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.12.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}