Andrew Shaw bearpelican

Building tools to democratize deep learning

bearpelican / Roman2Symbol_test_case.ipynb

Created February 28, 2019 23:26

Chord change after adding: `scale = [s+sec_offset for s in scale]`

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

bearpelican / medium_learn.py

Last active August 1, 2019 07:54

	# Data
	midi_files = get_files(midi_path, '.mid', recurse=True)
	data = MusicDataBunch.from_files(midi_files, data_path, processors=[Midi2ItemProcessor()])

	# Model
	learn = music_model_learner(data, arch=TransformerXL, config=default_config())

	# Train
	learn.fit_one_cycle(4)

bearpelican / medium_train.ipynb

Created July 31, 2019 01:09

Sorry, something went wrong. Reload?

Sorry, we cannot display this file.

Sorry, this file is invalid so it cannot be displayed.

bearpelican / show_la_bamba_snippet.py

Last active August 1, 2019 05:45

	f = midi_path/'La Bamba - Ritchie Valen - Chorus.mid'
	item = MusicItem.from_file(f, data.vocab)
	seed_item = item.trim_to_beat(6); seed_item.show()

bearpelican / bamba_prediction.py

Created August 1, 2019 07:32

	pitch_temp, dur_temp = (1.1, 0.4) # Controls amount of prediction randomness
	pred, full = learn.predict(seed_item, n_words=400, temperatures=(pitch_temp, dur_temp),
	top_k=24, top_p=0.7)
	full.show()

bearpelican / ChordProgression-I-V-vi.py

Created August 1, 2019 08:24

	# Let's create a partial progression I-V-vi
	p = music21.stream.Part()
	p.append(music21.chord.Chord('C4 E4 G4', type='half')) # I
	p.append(music21.chord.Chord('G3 B3 D4', type='half')) # V
	p.append(music21.chord.Chord('A3 C4 E4', type='half')) # vi
	s = music21.stream.Score([p])
	chord_item = MusicItem.from_stream(s, vocab)
	chord_item.show()

bearpelican / predict_I-V-vi.py

Created August 1, 2019 08:27

	_, full = learn.predict(chord_item, n_words=10, temperatures=(0.5,0.5))
	full.show()

bearpelican / multitransformer_snippet.py

Created August 1, 2019 10:29

	class MultiTransformer(nn.Module):
	"Multitask Transformer for training mask, next word, and sequence 2 sequence"
	def __init__(self, encoder, decoder, head, mem_len):
	super().__init__()
	self.encoder = encoder
	self.decoder = decoder
	self.head = head
	self.default_mem_len = mem_len
	self.current_mem_len = None

bearpelican / tokenize.py

Created August 2, 2019 03:29

	item = MusicItem.from_file('data/single_bar_example.mid', MusicVocab.create())
	item.to_text()

bearpelican / single_bar_example.txt

Created August 2, 2019 05:35

	<MidiFile 2 tracks
	<MidiTrack 0 -- 44 events
	<MidiEvent DeltaTime, t=0, track=0, channel=None>
	<MidiEvent SEQUENCE_TRACK_NAME, t=None, track=0, channel=None, data=b'Piano'>
	<MidiEvent DeltaTime, t=0, track=0, channel=None>
	<MidiEvent PROGRAM_CHANGE, t=None, track=0, channel=1, data=0>
	<MidiEvent DeltaTime, t=0, track=0, channel=None>
	<MidiEvent PITCH_BEND, t=None, track=0, channel=1, parameter1=0, parameter2=64>
	<MidiEvent DeltaTime, t=0, track=0, channel=None>
	<MidiEvent PROGRAM_CHANGE, t=None, track=0, channel=1, data=0>