def __init__(
self,
merge_mode: CategoricalValue("sum", "mul", "concat", "ave"),
units: DiscreteValue(32, 1024),
activation_fn: CategoricalValue("tanh", "sigmoid", "relu",
"linear"),
recurrent_activation_fn: CategoricalValue("tanh", "sigmoid",
"relu", "linear"),
dropout: ContinuousValue(0, 0.5),
recurrent_dropout: ContinuousValue(0, 0.5),
):
super().__init__(
layer=_LSTM(
units=units,
activation=activation_fn,
recurrent_activation=recurrent_activation_fn,
dropout=dropout,
recurrent_dropout=recurrent_dropout,
return_sequences=True,
),
merge_mode=merge_mode,
)
self.activation_fn = activation_fn
self.recurrent_activation_fn = recurrent_activation_fn
def __init__(
self,
featurewise_center: BooleanValue(),
samplewise_center: BooleanValue(),
featurewise_std_normalization: BooleanValue(),
samplewise_std_normalization: BooleanValue(),
rotation_range: DiscreteValue(0, 15),
width_shift_range: ContinuousValue(0, 0.25),
height_shift_range: ContinuousValue(0, 0.25),
shear_range: ContinuousValue(0, 15),
zoom_range: ContinuousValue(0, 0.25),
horizontal_flip: BooleanValue(),
vertical_flip: BooleanValue(),
):
super().__init__(
featurewise_center=featurewise_center,
samplewise_center=samplewise_center,
featurewise_std_normalization=featurewise_std_normalization,
samplewise_std_normalization=samplewise_std_normalization,
rotation_range=rotation_range,
width_shift_range=width_shift_range,
height_shift_range=height_shift_range,
shear_range=shear_range,
zoom_range=zoom_range,
horizontal_flip=horizontal_flip,
vertical_flip=vertical_flip,
)
def __init__(self, filters: DiscreteValue(2, 8),
kernel_size: CategoricalValue(3, 5, 7),
l1: ContinuousValue(0, 1e-3), l2: ContinuousValue(0, 1e-3),
**kwargs):
self.l1 = l1
self.l2 = l2
super().__init__(filters=2**filters,
kernel_size=(kernel_size, kernel_size),
kernel_regularizer=regularizers.l1_l2(l1=l1, l2=l2),
padding="same",
data_format="channels_last",
**kwargs)
def __init__(
self,
dm: DiscreteValue(min=0, max=2),
dbow_words: DiscreteValue(min=-100, max=100),
dm_concat: DiscreteValue(min=-100, max=100),
dm_tag_count: DiscreteValue(min=0, max=2),
alpha: ContinuousValue(min=0.001, max=0.075),
epochs: DiscreteValue(min=2, max=10),
window: DiscreteValue(min=2, max=10),
inner_tokenizer: algorithm(Sentence, Seq[Word]),
inner_stemmer: algorithm(Word, Stem),
inner_stopwords: algorithm(Seq[Word], Seq[Word]),
lowercase: BooleanValue(),
stopwords_remove: BooleanValue(),
):
self.inner_tokenizer = inner_tokenizer
self.inner_stemmer = inner_stemmer
self.inner_stopwords = inner_stopwords
self.lowercase = lowercase
self.stopwords_remove = stopwords_remove
super().__init__(
dm=dm,
dbow_words=dbow_words,
dm_concat=dm_concat,
dm_tag_count=dm_tag_count,
alpha=alpha,
epochs=epochs,
window=window,
)
def __init__(self, units: DiscreteValue(32, 1024),
activation_fn: CategoricalValue("tanh", "sigmoid", "relu",
"linear"),
recurrent_activation_fn: CategoricalValue(
"tanh", "sigmoid", "relu",
"linear"), dropout: ContinuousValue(0, 0.5),
recurrent_dropout: ContinuousValue(0, 0.5), **kwargs):
super().__init__(units=units,
activation=activation_fn,
recurrent_activation=recurrent_activation_fn,
dropout=dropout,
recurrent_dropout=recurrent_dropout,
return_sequences=False,
**kwargs)
self.activation_fn = activation_fn
self.recurrent_activation_fn = recurrent_activation_fn
def _get_arg_values(arg, value, cls):
if isinstance(value, bool):
return BooleanValue()
if isinstance(value, int):
return DiscreteValue(*_get_integer_values(arg, value, cls))
if isinstance(value, float):
return ContinuousValue(*_get_float_values(arg, value, cls))
if isinstance(value, str):
values = _find_parameter_values(arg, cls)
return CategoricalValue(*values) if values else None
return None
def __init__(
self,
dm: DiscreteValue(min=0, max=2),
dbow_words: DiscreteValue(min=-100, max=100),
dm_concat: DiscreteValue(min=-100, max=100),
dm_tag_count: DiscreteValue(min=0, max=2),
alpha: ContinuousValue(min=0.001, max=0.075),
epochs: DiscreteValue(min=2, max=10),
window: DiscreteValue(min=2, max=10),
):
self.dm = int(dm)
self.dbow_words = int(dbow_words)
self.dm_concat = int(dm_concat)
self.dm_tag_count = int(dm_tag_count)
self.alpha = alpha
self.epochs = int(epochs)
self.window = int(window)
def _get_float_values(arg, value, cls):
if value in [inf, nan]:
return None
if value > 0:
min_value = -10 * value
max_value = 10 * value
elif value == 0:
min_value = -1
max_value = 1
else:
return None
# binary search for minimum value
left = min_value
right = value
while abs(left - right) > 1e-2:
current_value = round((left + right) / 2, 3)
if _try(cls, arg, current_value):
right = current_value
else:
left = current_value
min_value = right
# binary search for maximum value
left = value
right = max_value
while abs(left - right) > 1e-2:
current_value = round((left + right) / 2, 3)
if _try(cls, arg, current_value):
left = current_value
else:
right = current_value
max_value = left
if max_value - min_value >= 2 * value:
return ContinuousValue(min=min_value, max=max_value)
return None
def __init__(self, rate: ContinuousValue(0, 0.5), **kwargs):
super().__init__(rate=rate, **kwargs)
def __init__(self, rate: ContinuousValue(0, 0.5)):
super().__init__(rate=rate)
def __init__(self, penalty: CategoricalValue("l1", "l2"),
C: ContinuousValue(0.1, 10)):
super().__init__(penalty=penalty, C=C, solver="liblinear")
def __init__(self, var_smoothing: ContinuousValue(1e-10, 0.1)):
super().__init__(var_smoothing=var_smoothing)
def __init__(
self,
kernel: CategoricalValue("rbf", "linear", "poly"),
C: ContinuousValue(0.1, 10),
):
super().__init__(C=C, kernel=kernel)