Create drum_parser_2column_data.py

w5/drum_parser_2column_data.py

@@ -0,0 +1,49 @@
+import music21
+
+o = music21.converter.parse('TheDanceOfEternity_dr.mid')
+
+dr_stream = music21.stream.Stream()
+drum_data = []
+
+for n in o.parts[0].getElementsByClass(music21.chord.Chord):
+    for p in n.pitches:
+        drum_data.append([p.midi, n.offset])
+
+for n in o.parts[0].getElementsByClass(music21.note.Note):
+    drum_data.append([n.pitch.midi, n.offset])
+
+
+import numpy as np
+
+drum_data = np.array(drum_data)
+drum_data = drum_data[drum_data[:, 1].argsort()]
+
+drum_data_diff = []
+drum_data_diff.append([drum_data[0,0], 0])
+
+for n in range(0, len(drum_data) - 2):
+    drum_data_diff.append([drum_data[n+1, 0], drum_data[n+1,1] - drum_data[n,1]])
+
+drum_data_diff = np.array(drum_data_diff)
+drum_note_set = set(drum_data_diff[:,0])
+drum_timing_diff_set = set(drum_data_diff[:,1])
+
+drum_data_diff_1d = []
+for n in range(0, len(drum_data_diff) - 1):
+    drum_data_diff_1d.append(drum_data_diff[n,0] + drum_data_diff[n,1] * 1j)
+
+drum_note_set = list(set(drum_data_diff_1d))
+drum_note_dic = {w: i for i, w in enumerate(drum_data_diff_1d)}
+
+import tensorflow as tf
+import numpy as np
+from tensorflow.contrib import rnn
+
+tf.set_random_seed(777)  # reproducibility
+
+data_dim = len(drum_note_set)
+hidden_size = len(drum_note_set)
+num_classes = len(drum_note_set)
+sequence_length = 100  # Any arbitrary number
+learning_rate = 0.1
+