forked from magenta/magenta
-
Notifications
You must be signed in to change notification settings - Fork 0
/
improv_rnn_pipeline.py
106 lines (91 loc) · 3.83 KB
/
improv_rnn_pipeline.py
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
# Copyright 2018 Google Inc. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Pipeline to create ImprovRNN dataset."""
import tensorflow as tf
import magenta
from magenta.pipelines import dag_pipeline
from magenta.pipelines import lead_sheet_pipelines
from magenta.pipelines import note_sequence_pipelines
from magenta.pipelines import pipeline
from magenta.pipelines import pipelines_common
from magenta.pipelines import statistics
from magenta.protobuf import music_pb2
class EncoderPipeline(pipeline.Pipeline):
"""A Module that converts lead sheets to a model specific encoding."""
def __init__(self, config, name):
"""Constructs an EncoderPipeline.
Args:
config: An ImprovRnnConfig that specifies the encoder/decoder,
pitch range, and transposition behavior.
name: A unique pipeline name.
"""
super(EncoderPipeline, self).__init__(
input_type=magenta.music.LeadSheet,
output_type=tf.train.SequenceExample,
name=name)
self._conditional_encoder_decoder = config.encoder_decoder
self._min_note = config.min_note
self._max_note = config.max_note
self._transpose_to_key = config.transpose_to_key
def transform(self, lead_sheet):
lead_sheet.squash(
self._min_note,
self._max_note,
self._transpose_to_key)
try:
encoded = [self._conditional_encoder_decoder.encode(
lead_sheet.chords, lead_sheet.melody)]
stats = []
except magenta.music.ChordEncodingException as e:
tf.logging.warning('Skipped lead sheet: %s', e)
encoded = []
stats = [statistics.Counter('chord_encoding_exception', 1)]
except magenta.music.ChordSymbolException as e:
tf.logging.warning('Skipped lead sheet: %s', e)
encoded = []
stats = [statistics.Counter('chord_symbol_exception', 1)]
self._set_stats(stats)
return encoded
def get_stats(self):
return {}
def get_pipeline(config, eval_ratio):
"""Returns the Pipeline instance which creates the RNN dataset.
Args:
config: An ImprovRnnConfig object.
eval_ratio: Fraction of input to set aside for evaluation set.
Returns:
A pipeline.Pipeline instance.
"""
all_transpositions = config.transpose_to_key is None
partitioner = pipelines_common.RandomPartition(
music_pb2.NoteSequence,
['eval_lead_sheets', 'training_lead_sheets'],
[eval_ratio])
dag = {partitioner: dag_pipeline.DagInput(music_pb2.NoteSequence)}
for mode in ['eval', 'training']:
time_change_splitter = note_sequence_pipelines.TimeChangeSplitter(
name='TimeChangeSplitter_' + mode)
quantizer = note_sequence_pipelines.Quantizer(
steps_per_quarter=config.steps_per_quarter, name='Quantizer_' + mode)
lead_sheet_extractor = lead_sheet_pipelines.LeadSheetExtractor(
min_bars=7, max_steps=512, min_unique_pitches=3, gap_bars=1.0,
ignore_polyphonic_notes=False, all_transpositions=all_transpositions,
name='LeadSheetExtractor_' + mode)
encoder_pipeline = EncoderPipeline(config, name='EncoderPipeline_' + mode)
dag[time_change_splitter] = partitioner[mode + '_lead_sheets']
dag[quantizer] = time_change_splitter
dag[lead_sheet_extractor] = quantizer
dag[encoder_pipeline] = lead_sheet_extractor
dag[dag_pipeline.DagOutput(mode + '_lead_sheets')] = encoder_pipeline
return dag_pipeline.DAGPipeline(dag)