class CorexText_Hyperparams(hyperparams.Hyperparams):
# number of Corex latent factors
n_hidden = UniformInt(
lower=1,
upper=301,
default=30,
description='number of topics',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
#
threshold = Uniform(
lower=0,
upper=10000,
default=0,
q=1,
description=
'threshold for number of columns in the tfidf matrix below which we don`t call CorEx',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
#
n_grams = UniformInt(
lower=1,
upper=10,
default=1,
description=
'n_grams parameter to use before feeding in text to TfidfVectorizer',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
# max_df @ http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfVectorizer.html
max_df = Uniform(
lower=0.0,
upper=1.00,
default=.9,
q=.05,
description='max percent document frequency of analysed terms',
semantic_types=[
"http://schema.org/Float",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
# min_df @ http://scikit-learn.org/stable/modules/generated/sklearn.feature_extraction.text.TfidfVectorizer.html
min_df = Uniform(
lower=0.0,
upper=1.00,
default=.02,
q=.01,
description='min percent document frequency of analysed terms',
semantic_types=[
"http://schema.org/Float",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
class CorexSAE_Hyperparams(hyperparams.Hyperparams):
label_beta = Uniform(
lower=0,
upper=1000,
default=1,
q=.01,
description=
'Lagrange multiplier for beta : 1 tradeoff btwn label relevance : compression.'
)
epochs = Uniform(lower=1,
upper=1000,
default=100,
description='number of epochs to train')
class SM_Hyperparams(hyperparams.Hyperparams):
reg_val = Uniform(
lower=0,
upper=1e-2,
q=1e-3,
default=1e-4,
description='l2 regularization penalty',
semantic_types=[
'http://schema.org/Float',
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
class CorexText_Hyperparams(hyperparams.Hyperparams):
n_hidden = Uniform(lower=0,
upper=100,
default=10,
q=1,
description='number of topics')
max_df = Uniform(
lower=.10,
upper=1.01,
default=.9,
q=.05,
description='max percent document frequency of analysed terms')
min_df = Union(OrderedDict([
('int df',
Uniform(
lower=1,
upper=20,
default=2,
q=1,
description='min integer document frequency of analysed terms')),
('pct df',
Uniform(
lower=0,
upper=.10,
default=.02,
q=.01,
description='min percent document frequency of analysed terms'))
]),
default='pct df')
chunking = Uniform(
lower=0,
upper=2000,
default=0,
q=100,
description=
'number of tfidf-filtered terms to include as a document, 0 => no chunking. last chunk may be > param value to avoid small documents'
)
max_features = Union(OrderedDict([
('none', Enumeration([None], default=None)),
('int mf',
Uniform(lower=1000,
upper=50001,
default=50000,
q=1000,
description='max number of terms to use'))
]),
default='none')
class EchoRegressor_Hyperparams(hyperparams.Hyperparams):
# regularization strength
alpha = Uniform(
lower=0,
upper=10,
default=1,
q=.1,
description='regularization strength',
semantic_types=[
"http://schema.org/Float",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
#
diagonal = UniformBool(
default=False,
description=
'assume diagonal covariance, leading to sparsity in data basis (instead of covariance eigenbasis)',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
class EchoIB_Hyperparams(hyperparams.Hyperparams):
n_hidden = UniformInt(
lower=1,
upper=401,
default=200,
description='number of hidden factors learned',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
beta = Uniform(
lower=0,
upper=1000,
default=.1,
q=.01,
description=
'Lagrange multiplier for beta (applied to regularizer I(X:Z)): defining tradeoff btwn label relevance : compression.',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
epochs = UniformInt(
lower=1,
upper=10000,
default=100,
description='number of epochs to train',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
batch = UniformInt(
lower=10,
upper=1000,
default=50,
description='batch_size',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
lr = LogUniform(
lower=0.00001,
upper=0.101,
default=0.001,
description='learning rate for Adam optimization',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
activation = Enumeration(
values=['relu', 'tanh', 'elu'],
default='tanh',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/ControlParameter'
],
description="activation to use for intermediate activations")
convolutional = UniformBool(
default=False,
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/ControlParameter'
],
description="whether to use a convolutional architecture")
task = Enumeration(
values=['CLASSIFICATION', 'REGRESSION'],
default='CLASSIFICATION',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/ControlParameter'
],
description='task type')
use_as_modeling = UniformBool(
default=False,
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/ControlParameter'
],
description=
"whether to return constructed features AND predictions (else, used for modeling i.e. only predictions"
)
units = UniformInt(
lower=10,
upper=401,
default=200,
description='# neurons in FC intermediate layers',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
layers = UniformInt(
lower=1,
upper=8,
default=2,
description='# of layers',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
error_on_no_input = hyperparams.UniformBool(
default=True,
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/ControlParameter'
],
description=
"Throw an exception if no input column is selected/provided. Defaults to true to behave like sklearn. To prevent pipelines from breaking set this to False."
)
gpus = Uniform(
lower=0,
upper=5,
q=1,
default=1,
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/ResourcesUseParameter'
],
description='GPUs to Use')
class SDNE_Hyperparams(hyperparams.Hyperparams):
dimension = UniformInt(
lower=10,
upper=200,
default=10,
#q = 5,
description='dimension of latent embedding',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
epochs = UniformInt(
lower=1,
upper=500,
default=50,
#q = 5e-8,
description='number of epochs to train',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
beta = UniformInt(
lower=1,
upper=20,
default=5,
#q = 1,
description=
'seen edge reconstruction weight (to account for sparsity in links for reconstructing adjacency. matrix B in Wang et al 2016',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
alpha = Uniform(
lower=1e-8,
upper=1,
default=1e-5,
#q = 5e-8,
description='first order proximity weight',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
lr = Uniform(
lower=1e-5,
upper=1e-2,
default=5e-4,
#q = 5e-8,
description='learning rate (constant across training)',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])
depth = UniformInt(
lower=1,
upper=10,
default=3,
#q = 5,
description='number of hidden layers',
semantic_types=[
"http://schema.org/Integer",
'https://metadata.datadrivendiscovery.org/types/ControlParameter'
])
return_list = UniformBool(
default=False,
description='for testing',
semantic_types=[
'https://metadata.datadrivendiscovery.org/types/TuningParameter'
])