RegularNPArrCache()
]
### Load Data
mnist_retrieval = MNIST_retrieval()
mnist_ds = DataSource(mnist_retrieval,
ImageFileDataEntity,
controllers=controllers)
train_ds, temp = DataSource.split(mnist_ds, split_percent=.6)
val_ds, test_ds = DataSource.split(temp, split_percent=.6)
### Load network
net = mnist_net(input_shape=(28, 28, 1), target_shape=10)
### Callbacks
callbacks = []
### Create and run experiment
exp = SimpleExperiment(train_datasource=train_ds,
validation_datasource=val_ds,
test_datasource=test_ds,
network=net,
metrics=['mae', 'acc'],
loss=categorical_crossentropy,
optimizer=Adadelta(),
label_names=list(range(10)),
callbacks=callbacks,
workers=3,
epochs=10)
exp.run()
val_ds, test_ds = DataSource.split(temp,
split_percent=.6) # Val at .24, test at .16
### Load network
## Changing to classification task. 88 notes x 11 note types(durations) x 12 relative onset times
'''
Note types: Whole, half, quarter, eighth, sixteenth, third, sixth, seventh, dotted Whole, dotted half, dotted quarter
Onset Times: Note types + 0 onset(chord)
'''
net = midi_test(input_shape=(100, 88, 12, 11), target_shape=(5, 88, 12, 11))
### Callbacks
callbacks = [ModelCheckpoint('weights.{epoch:02d}-{val_loss:.2f}.hdf5')]
### Create and run experiment
exp = SimpleExperiment(train_datasource=train_ds,
validation_datasource=val_ds,
test_datasource=test_ds,
callbacks=callbacks,
network=net,
metrics=['mae', 'acc'],
loss=categorical_crossentropy,
optimizer=Adadelta(),
epochs=1000,
workers=3,
use_multiprocessing=True,
train_batch_steps=10,
val_batch_steps=5)
exp.run()
batch_size=2,
verbosity=3,
controllers=controllers,
prefilters=prefilters)
val_ds = DataSource(val_retrieval,
ARFFDataEntity,
ignore_cache=True,
batch_size=2,
verbosity=3,
controllers=controllers,
prefilters=prefilters)
pdb.set_trace()
### Callbacks
callbacks = [ModelCheckpoint('weights.{epoch:02d}-{val_loss:.2f}.hdf5')]
### Network
net = dense_network(input_shape=(989, ), target_shape=(3))
### Create and run Experiment
exp = SimpleExperiment(train_datasource=train_ds,
validation_datasource=val_ds,
loss=KLD,
optimizer=Adadelta(),
metrics=['mae'],
network=net,
callbacks=callbacks)
exp.run()
#### VALIDATION DATASOURCE
iceberg_valds = DataSource(
val_retrieval,
ImageFileDataEntity,
controllers=[partial(accept_label, labelnames="is_iceberg")] + controllers,
**universal_options)
boat_valds = DataSource(
val_retrieval,
ImageFileDataEntity,
controllers=[partial(ignore_label, labelnames="is_iceberg")] + controllers,
**universal_options)
### Combining DataSources
train_ds = iceberg_trainds + boat_trainds
val_ds = iceberg_valds + boat_valds
#### Callbacks
callbacks = [ModelCheckpoint('weights.{epoch:02d}-{val_loss:.2f}.hdf5')]
#### Define experiment
exp = SimpleExperiment(train_datasource=train_ds,
validation_datasource=val_ds,
loss=categorical_crossentropy,
optimizer=Adadelta(),
network=net,
metrics=['acc'],
callbacks=callbacks)
exp.run()