'hidden_units': 2**(9 + hp.randint('hidden_units_log2', 3))
}
# Fixed parameters
search_space.update({
'max_layers': n_layers,
'lr_decay': 0.5,
'lr_decay_epocs': 250,
'n_threads': 4,
'memory_factor': 2,
'max_epochs': MAX_EPOCHS,
'strip_length': 5,
'progress_thresh': 0.1,
})
stats = tune_model(dataset=datasets.Datasets.COVERTYPE,
settings_fn=datasets.CoverTypeSettings,
search_space=search_space,
n_trials=CV_TRIALS,
cross_validate=False,
folder='cover',
runs=SIM_RUNS,
test_batch_size=1,
fine_tune=FineTuningType.ExtraLayerwise(
epochs_per_layer=20, policy=CyclicPolicy))
metrics = stats[0].keys()
for m in metrics:
values = [x[m] for x in stats]
logger.info('%s: %f +- %f' % (m, np.mean(values), np.std(values)))
'batch_norm': False,
'padding': 'VALID',
'max_layers': n_layers, # These are cnn layers here
'layerwise_progress_thresh': 0.1,
'n_threads': 4,
'memory_factor': 2,
'max_epochs': MAX_EPOCHS,
'tune_folder': 'incremental_2l_kernel',
'strip_length': 5,
'progress_thresh': 0.1,
'network_fn': cnn_kernel_example_layout_fn,
'policy': {'switch_policy': CyclicPolicy},
'fc_layers': 2
})
stats = tune_model(
dataset=datasets.Datasets.FASHION_MNIST,
settings_fn=datasets.FashionMnistSettings,
search_space=search_space,
n_trials=CV_TRIALS,
cross_validate=False,
folder='incremental_2l_kernel_stats',
runs=SIM_RUNS,
test_batch_size=128
)
metrics = stats[0].keys()
for m in metrics:
values = [x[m] for x in stats]
logger.info('%s: %f +- %f' % (m, np.mean(values), np.std(values)))
'padding': 'VALID',
'max_layers': n_layers, # These are cnn layers here
'layerwise_progress_thresh': 0.1,
'n_threads': 4,
'memory_factor': 2,
'max_epochs': MAX_EPOCHS,
'strip_length': 5,
'kernel_dropout_rate': 0.95,
'tune_folder': name + '_validate',
'progress_thresh': 0.1,
'network_fn': cnn_kernel_example_layout_fn,
'policy': {'switch_policy': CyclicPolicy},
'fc_layers': 2
})
stats = tune_model(
dataset=datasets.Datasets.CIFAR10,
settings_fn=datasets.Cifar10Settings,
search_space=search_space,
n_trials=CV_TRIALS,
cross_validate=False,
folder=name + '_stats',
runs=SIM_RUNS,
test_batch_size=128
)
metrics = stats[0].keys()
for m in metrics:
values = [x[m] for x in stats]
logger.info('%s: %f +- %f' % (m, np.mean(values), np.std(values)))
'hidden_units': 2**(6 + hp.randint('hidden_units_log2', 3))
}
# Fixed parameters
search_space.update({
'max_layers': n_layers,
'lr_decay': 0.5,
'lr_decay_epocs': 250,
'n_threads': 4,
'strip_length': 5,
'memory_factor': 1,
'max_epochs': MAX_EPOCHS,
'progress_thresh': 0.1,
})
stats = tune_model(dataset=datasets.Datasets.TITANIC,
settings_fn=datasets.TitanicSettings,
search_space=search_space,
n_trials=CV_TRIALS,
cross_validate=True,
folder='titanic',
runs=SIM_RUNS,
test_batch_size=1,
fine_tune=FineTuningType.ExtraLayerwise(
epochs_per_layer=20, policy=CyclicPolicy))
metrics = stats[0].keys()
for m in metrics:
values = [x[m] for x in stats]
logger.info('%s: %f +- %f' % (m, np.mean(values), np.std(values)))
# Fixed parameters
search_space.update({
'max_layers': n_layers,
'n_threads': 4,
'memory_factor': 2,
'batch_norm': False,
'max_epochs': MAX_EPOCHS,
'strip_length': 5,
'progress_thresh': 0.1,
'network_fn': kernel_example_layout_fn
})
stats = tune_model(
dataset=datasets.Datasets.MOTOR,
settings_fn=datasets.MotorSettings,
search_space=search_space,
n_trials=CV_TRIALS,
cross_validate=False,
folder='motor',
runs=SIM_RUNS,
test_batch_size=1,
#fine_tune=FineTuningType.ExtraLayerwise(
# epochs_per_layer=20, policy=CyclicPolicy
#)
)
metrics = stats[0].keys()
for m in metrics:
values = [x[m] for x in stats]
logger.info('%s: %f +- %f' % (m, np.mean(values), np.std(values)))
