def get_pyrbm(n_features, rng):
pyvotune.dense_input(PyRBMFeatureExtractor)
pyvotune.non_terminal(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='learning_rate',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='momentum',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='l2_weight',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='sparsity',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='scale',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pbool(name='binary', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pbool(name='reconstruction', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(1, 10), name='n_training_epochs',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(1, 20), name='n_gibbs',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(1, 1000), name='batch_size',
rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(5, 1500), name='n_hidden',
rng=rng)(PyRBMFeatureExtractor)
return [PyRBMFeatureExtractor]
def get_decomposers(n_features, rng):
pyvotune.dense_input(PCA)
pyvotune.non_terminal(PCA)
pyvotune.pbool(name='whiten', rng=rng)(PCA)
pyvotune.non_terminal(ProjectedGradientNMF)
pyvotune.pint(range=(1, n_features), name='n_components',
rng=rng)(ProjectedGradientNMF)
pyvotune.choice(choices=['nndsvd', 'nndsvda', 'nndsvdar'],
name='init',
rng=rng)(ProjectedGradientNMF)
pyvotune.choice(choices=['data', 'components', None],
name='sparseness',
rng=rng)(ProjectedGradientNMF)
pyvotune.pfloat(range=(1, 10), name='beta', rng=rng)(ProjectedGradientNMF)
pyvotune.pfloat(range=(0.01, 10), name='eta',
rng=rng)(ProjectedGradientNMF)
pyvotune.pint(range=(1, 600), name='max_iter',
rng=rng)(ProjectedGradientNMF)
pyvotune.pint(range=(100, 4000), name='nls_max_iter',
rng=rng)(ProjectedGradientNMF)
# PCA & ProbabilisticPCA are initialized together since ProbabilisticPCA
# wraps PCA
return [PCA, ProbabilisticPCA, ProjectedGradientNMF]
def get_decomposers(n_features, rng):
pyvotune.dense_input(PCA)
pyvotune.non_terminal(PCA)
pyvotune.pbool(name='whiten', rng=rng)(PCA)
pyvotune.non_terminal(ProjectedGradientNMF)
pyvotune.pint(
range=(1, n_features), name='n_components', rng=rng)(ProjectedGradientNMF)
pyvotune.choice(
choices=['nndsvd', 'nndsvda', 'nndsvdar'],
name='init', rng=rng)(ProjectedGradientNMF)
pyvotune.choice(
choices=['data', 'components', None],
name='sparseness', rng=rng)(ProjectedGradientNMF)
pyvotune.pfloat(
range=(1, 10), name='beta', rng=rng)(ProjectedGradientNMF)
pyvotune.pfloat(
range=(0.01, 10), name='eta', rng=rng)(ProjectedGradientNMF)
pyvotune.pint(
range=(1, 600), name='max_iter', rng=rng)(ProjectedGradientNMF)
pyvotune.pint(
range=(100, 4000), name='nls_max_iter', rng=rng)(ProjectedGradientNMF)
# PCA & ProbabilisticPCA are initialized together since ProbabilisticPCA
# wraps PCA
return [PCA, ProbabilisticPCA, ProjectedGradientNMF]
def get_theano(n_features, rng):
pyvotune.dense_input(TheanoRBMFeatureExtractor)
pyvotune.non_terminal(TheanoRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='learning_rate', rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(1, 10), name='training_epochs', rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(1, 1000), name='batch_size', rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(1, 1000), name='n_resamples', rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(5, 1500), name='n_hidden', rng=rng)(TheanoRBMFeatureExtractor)
return [TheanoRBMFeatureExtractor]
def get_theano(n_features, rng):
pyvotune.dense_input(TheanoRBMFeatureExtractor)
pyvotune.non_terminal(TheanoRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='learning_rate',
rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(1, 10), name='training_epochs',
rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(1, 1000), name='batch_size',
rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(1, 1000), name='n_resamples',
rng=rng)(TheanoRBMFeatureExtractor)
pyvotune.pint(range=(5, 1500), name='n_hidden',
rng=rng)(TheanoRBMFeatureExtractor)
return [TheanoRBMFeatureExtractor]
def get_pyrbm(n_features, rng):
pyvotune.dense_input(PyRBMFeatureExtractor)
pyvotune.non_terminal(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='learning_rate', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='momentum', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='l2_weight', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='sparsity', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pfloat(range=(0., 1.0), name='scale', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pbool(name='binary', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pbool(name='reconstruction', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(1, 10), name='n_training_epochs', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(1, 20), name='n_gibbs', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(1, 1000), name='batch_size', rng=rng)(PyRBMFeatureExtractor)
pyvotune.pint(range=(5, 1500), name='n_hidden', rng=rng)(PyRBMFeatureExtractor)
return [PyRBMFeatureExtractor]
def get_image_features(n_features, rng):
pyvotune.dense_input(PatchExtractor)
pyvotune.non_terminal(PatchExtractor)
pyvotune.param(
typename="patchsize",
checker_fn=patchsize_checker,
checker_args={
'min_width': 2,
'max_width': 200,
'min_height': 2,
'max_height': 200
},
generator_fn=patchsize_generator,
generator_args={
'min_width': 2,
'max_width': 200,
'min_height': 2,
'max_height': 200
},
name="patch_size")(PatchExtractor)
pyvotune.pint(range=(1, 100), name='max_patches', rng=rng)(PatchExtractor)
return [PatchExtractor]
def get_preprocessors(n_features, rng):
pyvotune.dense_input(Scaler)
pyvotune.non_terminal(Scaler)
pyvotune.pbool(name="with_std", rng=rng)(Scaler)
pyvotune.non_terminal(Normalizer)
pyvotune.choice(choices=["l1", "l2"], name="norm", rng=rng)(Normalizer)
pyvotune.non_terminal(Binarizer)
pyvotune.pfloat(range=(0, 10000), name="threshold", rng=rng)(Binarizer)
return [Scaler, Normalizer, Binarizer]