will run on windows

Hiroshiba · Aug 24, 2019 · 9fd4ebd · 9fd4ebd
1 parent 83fccf4
commit 9fd4ebd
Show file tree

Hide file tree

Showing 6 changed files with 140 additions and 138 deletions.
diff --git a/become_yukarin/dataset/dataset.py b/become_yukarin/dataset/dataset.py
@@ -485,7 +485,7 @@ def create(config: DatasetConfig):
     # cropping
     if config.train_crop_size is not None:
         def add_seed():
-            return LambdaProcess(lambda d, test: dict(seed=numpy.random.randint(2 ** 32), **d))
+            return LambdaProcess(lambda d, test: dict(seed=numpy.random.randint(2 ** 31), **d))
 
         def padding(s):
             return ChainProcess([
@@ -587,7 +587,7 @@ def create_sr(config: SRDatasetConfig):
     # cropping
     if config.train_crop_size is not None:
         def add_seed():
-            return LambdaProcess(lambda d, test: dict(seed=numpy.random.randint(2 ** 32), **d))
+            return LambdaProcess(lambda d, test: dict(seed=numpy.random.randint(2 ** 31), **d))
 
         def padding(s):
             return ChainProcess([

diff --git a/recipe/config_sr.json b/recipe/config_sr.json
@@ -1,6 +1,6 @@
 {
   "dataset": {
-    "input_glob": "/yukari-news-harvest/yukari-news-*.npy",
+    "input_glob": "./feature/*.npy",
     "num_test": 5,
     "input_global_noise": 3,
     "input_local_noise": 3,

diff --git a/scripts/extract_acoustic_feature.py b/scripts/extract_acoustic_feature.py
@@ -46,8 +46,6 @@
 parser.add_argument('--enable_overwrite', action='store_true')
 arguments = parser.parse_args()
 
-pprint(dir(arguments))
-
 pre_convert = arguments.pre_converter1_config is not None
 if pre_convert:
     config = create_config(arguments.pre_converter1_config)
@@ -188,6 +186,8 @@ def concatenate(arr_list):
 
 
 def main():
+    pprint(vars(arguments))
+
     paths1 = list(sorted(arguments.input1_directory.glob('*')))
     paths2 = list(sorted(arguments.input2_directory.glob('*')))
     assert len(paths1) == len(paths2)

diff --git a/scripts/extract_spectrogram_pair.py b/scripts/extract_spectrogram_pair.py
@@ -33,8 +33,6 @@
 parser.add_argument('--enable_overwrite', action='store_true')
 arguments = parser.parse_args()
 
-pprint(dir(arguments))
-
 
 def generate_file(path):
     out = Path(arguments.output_directory, path.stem + '.npy')
@@ -74,6 +72,8 @@ def generate_file(path):
 
 
 def main():
+    pprint(vars(arguments))
+
     paths = list(sorted(arguments.input_directory.glob('*')))
     arguments.output_directory.mkdir(exist_ok=True)
 

diff --git a/train.py b/train.py
@@ -14,68 +14,69 @@
 from become_yukarin.model.model import create
 from become_yukarin.updater.updater import Updater
 
-parser = argparse.ArgumentParser()
-parser.add_argument('config_json_path', type=Path)
-parser.add_argument('output', type=Path)
-arguments = parser.parse_args()
-
-config = create_from_json(arguments.config_json_path)
-arguments.output.mkdir(exist_ok=True)
-config.save_as_json((arguments.output / 'config.json').absolute())
-
-# model
-if config.train.gpu >= 0:
-    cuda.get_device_from_id(config.train.gpu).use()
-predictor, discriminator = create(config.model)
-models = {
-    'predictor': predictor,
-    'discriminator': discriminator,
-}
-
-# dataset
-dataset = create_dataset(config.dataset)
-train_iter = MultiprocessIterator(dataset['train'], config.train.batchsize)
-test_iter = MultiprocessIterator(dataset['test'], config.train.batchsize, repeat=False, shuffle=False)
-train_eval_iter = MultiprocessIterator(dataset['train_eval'], config.train.batchsize, repeat=False, shuffle=False)
-
-
-# optimizer
-def create_optimizer(model):
-    optimizer = optimizers.Adam(alpha=0.0002, beta1=0.5, beta2=0.999)
-    optimizer.setup(model)
-    return optimizer
-
-
-opts = {key: create_optimizer(model) for key, model in models.items()}
-
-# updater
-converter = partial(convert.concat_examples, padding=0)
-updater = Updater(
-    loss_config=config.loss,
-    predictor=predictor,
-    discriminator=discriminator,
-    device=config.train.gpu,
-    iterator=train_iter,
-    optimizer=opts,
-    converter=converter,
-)
-
-# trainer
-trigger_log = (config.train.log_iteration, 'iteration')
-trigger_snapshot = (config.train.snapshot_iteration, 'iteration')
-
-trainer = training.Trainer(updater, out=arguments.output)
-
-ext = extensions.Evaluator(test_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
-trainer.extend(ext, name='test', trigger=trigger_log)
-ext = extensions.Evaluator(train_eval_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
-trainer.extend(ext, name='train', trigger=trigger_log)
-
-trainer.extend(extensions.dump_graph('predictor/loss'))
-
-ext = extensions.snapshot_object(predictor, filename='predictor_{.updater.iteration}.npz')
-trainer.extend(ext, trigger=trigger_snapshot)
-
-trainer.extend(extensions.LogReport(trigger=trigger_log))
-
-trainer.run()
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('config_json_path', type=Path)
+    parser.add_argument('output', type=Path)
+    arguments = parser.parse_args()
+
+    config = create_from_json(arguments.config_json_path)
+    arguments.output.mkdir(exist_ok=True)
+    config.save_as_json((arguments.output / 'config.json').absolute())
+
+    # model
+    if config.train.gpu >= 0:
+        cuda.get_device_from_id(config.train.gpu).use()
+    predictor, discriminator = create(config.model)
+    models = {
+        'predictor': predictor,
+        'discriminator': discriminator,
+    }
+
+    # dataset
+    dataset = create_dataset(config.dataset)
+    train_iter = MultiprocessIterator(dataset['train'], config.train.batchsize)
+    test_iter = MultiprocessIterator(dataset['test'], config.train.batchsize, repeat=False, shuffle=False)
+    train_eval_iter = MultiprocessIterator(dataset['train_eval'], config.train.batchsize, repeat=False, shuffle=False)
+
+
+    # optimizer
+    def create_optimizer(model):
+        optimizer = optimizers.Adam(alpha=0.0002, beta1=0.5, beta2=0.999)
+        optimizer.setup(model)
+        return optimizer
+
+
+    opts = {key: create_optimizer(model) for key, model in models.items()}
+
+    # updater
+    converter = partial(convert.concat_examples, padding=0)
+    updater = Updater(
+        loss_config=config.loss,
+        predictor=predictor,
+        discriminator=discriminator,
+        device=config.train.gpu,
+        iterator=train_iter,
+        optimizer=opts,
+        converter=converter,
+    )
+
+    # trainer
+    trigger_log = (config.train.log_iteration, 'iteration')
+    trigger_snapshot = (config.train.snapshot_iteration, 'iteration')
+
+    trainer = training.Trainer(updater, out=arguments.output)
+
+    ext = extensions.Evaluator(test_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
+    trainer.extend(ext, name='test', trigger=trigger_log)
+    ext = extensions.Evaluator(train_eval_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
+    trainer.extend(ext, name='train', trigger=trigger_log)
+
+    trainer.extend(extensions.dump_graph('predictor/loss'))
+
+    ext = extensions.snapshot_object(predictor, filename='predictor_{.updater.iteration}.npz')
+    trainer.extend(ext, trigger=trigger_snapshot)
+
+    trainer.extend(extensions.LogReport(trigger=trigger_log))
+
+    trainer.run()
diff --git a/train_sr.py b/train_sr.py
@@ -14,69 +14,70 @@
 from become_yukarin.model.sr_model import create_sr as create_sr_model
 from become_yukarin.updater.sr_updater import SRUpdater
 
-parser = argparse.ArgumentParser()
-parser.add_argument('config_json_path', type=Path)
-parser.add_argument('output', type=Path)
-arguments = parser.parse_args()
-
-config = create_from_json(arguments.config_json_path)
-arguments.output.mkdir(exist_ok=True)
-config.save_as_json((arguments.output / 'config.json').absolute())
-
-# model
-if config.train.gpu >= 0:
-    cuda.get_device_from_id(config.train.gpu).use()
-predictor, discriminator = create_sr_model(config.model)
-models = {
-    'predictor': predictor,
-    'discriminator': discriminator,
-}
-
-# dataset
-dataset = create_sr_dataset(config.dataset)
-train_iter = MultiprocessIterator(dataset['train'], config.train.batchsize)
-test_iter = MultiprocessIterator(dataset['test'], config.train.batchsize, repeat=False, shuffle=False)
-train_eval_iter = MultiprocessIterator(dataset['train_eval'], config.train.batchsize, repeat=False, shuffle=False)
-
-
-# optimizer
-def create_optimizer(model):
-    optimizer = optimizers.Adam(alpha=0.0002, beta1=0.5, beta2=0.999)
-    optimizer.setup(model)
-    return optimizer
-
-
-opts = {key: create_optimizer(model) for key, model in models.items()}
-
-# updater
-converter = partial(convert.concat_examples, padding=0)
-updater = SRUpdater(
-    loss_config=config.loss,
-    predictor=predictor,
-    discriminator=discriminator,
-    device=config.train.gpu,
-    iterator=train_iter,
-    optimizer=opts,
-    converter=converter,
-)
-
-# trainer
-trigger_log = (config.train.log_iteration, 'iteration')
-trigger_snapshot = (config.train.snapshot_iteration, 'iteration')
-
-trainer = training.Trainer(updater, out=arguments.output)
-
-ext = extensions.Evaluator(test_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
-trainer.extend(ext, name='test', trigger=trigger_log)
-ext = extensions.Evaluator(train_eval_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
-trainer.extend(ext, name='train', trigger=trigger_log)
-
-trainer.extend(extensions.dump_graph('predictor/loss'))
-
-ext = extensions.snapshot_object(predictor, filename='predictor_{.updater.iteration}.npz')
-trainer.extend(ext, trigger=trigger_snapshot)
-
-trainer.extend(extensions.LogReport(trigger=trigger_log))
-trainer.extend(extensions.PrintReport(['predictor/loss']))
-
-trainer.run()
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument('config_json_path', type=Path)
+    parser.add_argument('output', type=Path)
+    arguments = parser.parse_args()
+
+    config = create_from_json(arguments.config_json_path)
+    arguments.output.mkdir(exist_ok=True)
+    config.save_as_json((arguments.output / 'config.json').absolute())
+
+    # model
+    if config.train.gpu >= 0:
+        cuda.get_device_from_id(config.train.gpu).use()
+    predictor, discriminator = create_sr_model(config.model)
+    models = {
+        'predictor': predictor,
+        'discriminator': discriminator,
+    }
+
+    # dataset
+    dataset = create_sr_dataset(config.dataset)
+    train_iter = MultiprocessIterator(dataset['train'], config.train.batchsize)
+    test_iter = MultiprocessIterator(dataset['test'], config.train.batchsize, repeat=False, shuffle=False)
+    train_eval_iter = MultiprocessIterator(dataset['train_eval'], config.train.batchsize, repeat=False, shuffle=False)
+
+
+    # optimizer
+    def create_optimizer(model):
+        optimizer = optimizers.Adam(alpha=0.0002, beta1=0.5, beta2=0.999)
+        optimizer.setup(model)
+        return optimizer
+
+
+    opts = {key: create_optimizer(model) for key, model in models.items()}
+
+    # updater
+    converter = partial(convert.concat_examples, padding=0)
+    updater = SRUpdater(
+        loss_config=config.loss,
+        predictor=predictor,
+        discriminator=discriminator,
+        device=config.train.gpu,
+        iterator=train_iter,
+        optimizer=opts,
+        converter=converter,
+    )
+
+    # trainer
+    trigger_log = (config.train.log_iteration, 'iteration')
+    trigger_snapshot = (config.train.snapshot_iteration, 'iteration')
+
+    trainer = training.Trainer(updater, out=arguments.output)
+
+    ext = extensions.Evaluator(test_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
+    trainer.extend(ext, name='test', trigger=trigger_log)
+    ext = extensions.Evaluator(train_eval_iter, models, converter, device=config.train.gpu, eval_func=updater.forward)
+    trainer.extend(ext, name='train', trigger=trigger_log)
+
+    trainer.extend(extensions.dump_graph('predictor/loss'))
+
+    ext = extensions.snapshot_object(predictor, filename='predictor_{.updater.iteration}.npz')
+    trainer.extend(ext, trigger=trigger_snapshot)
+
+    trainer.extend(extensions.LogReport(trigger=trigger_log))
+    trainer.extend(extensions.PrintReport(['predictor/loss']))
+
+    trainer.run()