def create_hooks(self, config):
hooks = super().create_hooks(config)
#LOGGING HOOKS
# TODO here I am ignoring the number of channels, is it correct? (Luigi).
# TODO The channels seem to be included with np.prod...
# dim_with_channels = np.prod(self.x_shape["train"])
dim_with_channels = tf.cast(tf.reduce_prod(tf.shape(self.raw_x)[1:]),
tf.float32)
# TODO the two shapes could be different, maybe use tf.shape(self.raw_x) (Riccardo)
# dim_with_channels = np.prod(self.x_shape["train"])
# dim_with_channels = np.prod(self.x_shape["eval"])
# TODO I think this formula is still not correct, it is using loss (which inside has beta and warm_up), it is more correct than the one before maybe
# check https://www.reddit.com/r/MachineLearning/comments/56m5o2/discussion_calculation_of_bitsdims/
bits_dim = ((self.loss / dim_with_channels) +
tf.log(256.0 / 2.0)) / tf.log(2.0)
tensors_to_average = [[[-self.loss], [bits_dim]],
self.loss_nodes_to_log]
tensors_to_average_names = [[["LB_log(p)"], ["b/d"]],
self.loss_nodes_to_log_names]
tensors_to_average_plots = [[{
"fileName": "loss"
}, {
"fileName": "bits_dim"
}], self.loss_nodes_to_log_filenames]
#[*[name for name in self.loss_nodes_to_track]],
hooks.append(
LoggingMeanTensorsHook(
model=self,
fileName="log",
dirName=self.dirName,
tensors_to_average=tensors_to_average,
tensors_to_average_names=tensors_to_average_names,
tensors_to_average_plots=tensors_to_average_plots,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
# if you want to remove some dataset from here, make support to specify from conf on which datasets to log, if in doubt ask me please. Riccardo
datasets_keys=[TRAIN, VALIDATION, TEST],
time_reference=self._time_reference_str))
kwargs = config.get("ImagesReconstructHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
VAEImagesReconstructHook(model=self,
dirName=self.dirName,
**kwargs))
kwargs = config.get("ImagesGenerateHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
ImagesGenerateHook(model=self, dirName=self.dirName, **kwargs))
kwargs = config.get("ImportanceSamplingHook", None)
if kwargs:
# create the IS node only when needed, this function fails when z is fully convolutional... (Riccardo)
self.importance_sampling_node = self._create_importance_sampling_node(
)
if not isinstance(kwargs, list):
kwargs = [kwargs]
for kw in kwargs:
kws = {**self._plot_model_hooks_kwargs, **kw}
hooks.append(
ImportanceSamplingHook(
model=self,
dirName=self.dirName,
tensors_to_average=[self.importance_sampling_node],
datasets_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kws))
kwargs = config.get("TwoDimPCALatentVariablesHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
TwoDimPCALatentVariablesHook(
model=self,
dirName=self.dirName,
tensors=[self.z] +
list(self._approximate_posterior_params),
tensors_names=['z', 'mu'],
datasets_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kwargs))
kwargs = config.get("PCALatentVariablesHook", None)
if kwargs:
kwargs = {**self._plot_model_hooks_kwargs, **kwargs}
hooks.append(
PCALatentVariablesHook(
model=self,
dirName=self.dirName,
tensors=[self.z, self._approximate_posterior_params[0]],
tensors_names=['z', 'mu'],
datasets_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kwargs))
kwargs = config.get("VAELinearInterpolationHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
VAELinearInterpolationHook(model=self,
dirName=self.dirName,
**kwargs))
kwargs = config.get("LatentVarsClassificationHook", None)
if kwargs:
kwargs = {**self._plot_model_hooks_kwargs, **kwargs}
hooks.append(
LatentVarsClassificationHook(
model=self,
dirName=self.dirName,
tensors=[
self.z, self._approximate_posterior_params[0],
tf.concat(list(self._approximate_posterior_params),
axis=1)
],
tensors_names=['z', 'mu', 'mu_cov'],
datasets_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kwargs))
# frechet inception distance
kwargs = config.get("FrechetInceptionDistanceHook", None)
if kwargs:
for kw in kwargs:
kws = {**self._default_model_hooks_kwargs, **kw}
hooks.append(
FrechetInceptionDistanceHook(model=self,
dirName=self.dirName,
**kws))
# latent traversals hook
kwargs = config.get("LatentTraversalsHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
LatentTraversalsHook(model=self,
dirName=self.dirName,
**kwargs))
kwargs = config.get("LatentVarsGeometricClassificationHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
LatentVarsGeometricClassificationHook(
model=self,
dirName=self.dirName,
tensors=[
self.z, self._approximate_posterior_params[0],
tf.concat(list(self._approximate_posterior_params),
axis=1)
],
tensors_names=['z', 'mu', 'mu_cov'],
datasets_keys=[TRAIN, VALIDATION, TEST],
**kwargs))
return hooks
def create_hooks(self, config):
hooks = super().create_hooks(config)
tensors_to_average = [
[[self.cost]],
[[self._prior.mean()], [self.n_z_samples]],
]
tensors_to_average_names = [
[["loss_wake"]],
[["prior_mean"], ["samples"]],
]
tensors_to_average_plots = [[
{
"fileName": "loss"
},
], [{
"fileName": "mean"
}, {
"fileName": "samples"
}]]
hooks.append(
LoggingMeanTensorsHook(
model=self,
fileName="log",
dirName=self.dirName,
tensors_to_average=tensors_to_average,
tensors_to_average_names=tensors_to_average_names,
tensors_to_average_plots=tensors_to_average_plots,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
datasets_keys=[VALIDATION],
time_reference=self._time_reference_str))
tensors_to_average = [
self.loss_nodes_to_log,
]
tensors_to_average_names = [
self.loss_nodes_to_log_names,
]
tensors_to_average_plots = [
self.loss_nodes_to_log_filenames,
]
hooks.append(
LoggingMeanTensorsHook(
model=self,
fileName="log2",
dirName=self.dirName,
tensors_to_average=tensors_to_average,
tensors_to_average_names=tensors_to_average_names,
tensors_to_average_plots=tensors_to_average_plots,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
print_to_screen=False,
trigger_summaries=config["save_summaries"],
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
datasets_keys=[],
time_reference=self._time_reference_str))
kwargs = config.get("HMFisherMatrixHook2", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
HMFisherMatrixHook2(model=self, dirName=self.dirName,
**kwargs))
else:
if self._optimizer._d_p is not None:
t_to_average = [[[self._optimizer._diagonal_pad]]]
t_to_average_names = [
[["diagonal_pad"]],
]
t_to_average_plots = [[{"fileName": "diagonal_pad"}]]
hooks.append(
LoggingMeanTensorsHook(
model=self,
fileName="dp",
dirName=self.dirName,
tensors_to_average=t_to_average,
tensors_to_average_names=t_to_average_names,
tensors_to_average_plots=t_to_average_plots,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
print_to_screen=False,
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
datasets_keys=[]))
kwargs = config.get("ImagesGenerateHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
ImagesGenerateHook(model=self,
dirName=self.dirName,
pm_one=self._pm_one,
**kwargs))
kwargs = config.get("ImagesInputHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
ImagesInputHook(model=self,
dirName=self.dirName,
pm_one=self._pm_one,
**kwargs))
kwargs = config.get("ImagesReconstructHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
ImagesReconstructHook(model=self,
dirName=self.dirName,
pm_one=self._pm_one,
**kwargs))
kwargs = config.get("ThreeByThreeHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
ThreeByThreeHook(model=self,
tensorboard_dir=self._tensorboard_dir,
dirName=self.dirName,
**kwargs))
kwargs = config.get("LogpImportanceSamplingHook", None)
if kwargs:
if not isinstance(kwargs, list):
kwargs = [kwargs]
for kw in kwargs:
kws = {**self._plot_model_hooks_kwargs, **kw}
hooks.append(
ImportanceSamplingHook(
model=self,
dirName=self.dirName,
tensors_to_average=[self.importance_sampling_node],
datasets_keys=[TRAIN, VALIDATION, TEST],
**kws))
return hooks
def _create_gradient_hook(self, config):
# gradienthook
tensors_to_average = [
[
[self.gradient_weight_global_norms[PHASE_WAKE][0]],
self.gradient_norms[PHASE_WAKE],
],
[[self.gradient_weight_global_norms[PHASE_SLEEP][0]],
self.gradient_norms[PHASE_SLEEP]],
[
[self.gradient_weight_global_norms[PHASE_WAKE][1]],
self.weight_norms[PHASE_WAKE],
],
[[self.gradient_weight_global_norms[PHASE_SLEEP][1]],
self.weight_norms[PHASE_SLEEP]],
]
layer_names_wake = [
"L" + name.split(":")[0]
for name in self.gradient_names[PHASE_WAKE]
]
layer_names_sleep = [
"L" + name.split(":")[0]
for name in self.gradient_names[PHASE_SLEEP]
]
tensors_to_average_names = [
[
["gradient_global_norms_wake"],
layer_names_wake,
],
[["gradient_global_norms_sleep"], layer_names_sleep],
[
["weight_global_norms_wake"],
layer_names_wake,
],
[["weight_global_norms_sleep"], layer_names_sleep],
]
tensors_to_average_plots = [
[
{
"fileName": "gradient_global_norms_wake",
"logscale-y": 1,
"compose-label": 0
},
{
"fileName": "gradient_norms_wake",
"logscale-y": 1,
"compose-label": 0
},
],
[{
"fileName": "gradient_global_norms_sleep",
"logscale-y": 1,
"compose-label": 0
}, {
"fileName": "gradient_norms_sleep",
"logscale-y": 1,
"compose-label": 0
}],
[
{
"fileName": "weight_global_norms_wake",
"logscale-y": 1,
"compose-label": 0
},
{
"fileName": "weight_norms_wake",
"logscale-y": 1,
"compose-label": 0
},
],
[{
"fileName": "weight_global_norms_sleep",
"logscale-y": 1,
"compose-label": 0
}, {
"fileName": "weight_norms_sleep",
"logscale-y": 1,
"compose-label": 0
}],
]
kwargs = config.get("GradientsHook", None)
hook = None
if kwargs:
gradient_period = config["GradientsHook"]["period"]
gradient_steps = self._get_steps(gradient_period,
self._time_reference_str)
hook = LoggingMeanTensorsHook(
model=self,
fileName="gradient",
dirName=self.dirName,
tensors_to_average=tensors_to_average,
tensors_to_average_names=tensors_to_average_names,
tensors_to_average_plots=tensors_to_average_plots,
average_steps=gradient_steps,
time_reference=self._time_reference_str,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
# trigger_plot = True,
print_to_screen=False,
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
datasets_keys=[])
return hook
def create_hooks(self, config):
hooks = super().create_hooks(config)
# LOGGING HOOKS
# the shape of the input could be different from train to eval
dim_with_channels = tf.cast(tf.reduce_prod(tf.shape(self.x)[1:]),
tf.float32)
# check https://www.reddit.com/r/MachineLearning/comments/56m5o2/discussion_calculation_of_bitsdims/
bits_dim = (self.loss / dim_with_channels) / tf.log(
2.0) # - tf.log(256.0)
psnr_reconstruction_quality = tf.image.psnr(
self.x,
self.x_reconstruction_node_tf,
max_val=tf.reduce_max(self.x))
sample_rate = self.dataset.sample_rate
mfcc_original = mfcc(self.x, samplerate=sample_rate, preemph=0)
mfcc_reconstructed = mfcc(self.x_reconstruction_node_tf,
samplerate=sample_rate,
preemph=0)
mcd_reconstruction_quality = mcd(mfcc_original, mfcc_reconstructed)
tensors_to_average = [[[psnr_reconstruction_quality]],
[[mcd_reconstruction_quality]], [
[-self.loss],
], [
[bits_dim],
], self.nodes_to_track]
tensors_to_average_names = [[['PSNR_reconstruction_quality']],
[['MCD_reconstruction_distortion']],
[["LB_log(p)"]], [["b/d"]],
self.names_nodes_to_track]
tensors_to_average_plots = [[{
"fileName": 'psnr_reconstr_quality'
}], [{
"fileName": 'mcd_reconstr_distortion'
}], [{
"fileName": "loss"
}], [{
"fileName": "bits_dim"
}], self.loss_nodes_to_log_filenames]
hooks.append(
LoggingMeanTensorsHook(
model=self,
fileName="log",
dirName=self.dirName,
tensors_to_average=tensors_to_average,
tensors_to_average_names=tensors_to_average_names,
tensors_to_average_plots=tensors_to_average_plots,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
datasets_keys=[VALIDATION],
time_reference=self._time_reference_str))
kwargs = config.get("WavReconstructHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
WavReconstructHook(model=self, dirName=self.dirName, **kwargs))
kwargs = config.get("WavenetGaussianVisualizationHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
WavenetGaussianVisualizationHook(model=self,
dirName=self.dirName,
**kwargs))
kwargs = config.get("WavGenerateHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
WavGenerateHook(model=self, dirName=self.dirName, **kwargs))
kwargs = config.get("TwoDimPCALatentVariablesHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
TwoDimPCALatentVariablesHook(
model=self,
dirName=self.dirName,
tensors=[self.z],
tensors_names=['z'],
datasets_keys=[TRAIN, VALIDATION],
# don't change the order (Luigi)
**kwargs))
kwargs = config.get("PCALatentVariablesHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
PCALatentVariablesHook(
model=self,
dirName=self.dirName,
tensors=[tf.reshape(self.z, [-1, self.z.shape[-1]])],
tensors_names=['z'],
datasets_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kwargs))
kwargs = config.get("WavLatentPCAHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
WavLatentPCAHook(
model=self,
dirName=self.dirName,
dataset_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kwargs))
kwargs = config.get('WavClusterAnomalyDetectionHook', None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
WavClusterAnomalyDetectionHook(
model=self,
dirName=self.dirName,
dataset_keys=[TRAIN, VALIDATION
], # don't change the order (Luigi)
**kwargs))
return hooks
def create_hooks(self, config):
hooks = super().create_hooks(config)
# LOGGING HOOKS
tensors_to_average = self.loss_nodes_to_log
tensors_to_average_names = self.loss_nodes_to_log_names
tensors_to_average_plots = self.loss_nodes_to_log_filenames
hooks.append(LoggingMeanTensorsHook(model=self,
fileName="log",
dirName=self.dirName,
tensors_to_average=tensors_to_average,
tensors_to_average_names=tensors_to_average_names,
tensors_to_average_plots=tensors_to_average_plots,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
# if you want to remove some dataset from here, make support to specify from conf on which datasets to log, if in doubt ask me please. Riccardo
datasets_keys=[TRAIN, VALIDATION, TEST],
time_reference=self._time_reference_str
)
)
kwargs = config.get("ImagesReconstructHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs,
**kwargs}
hooks.append(AEImagesReconstructHook(model=self,
dirName=self.dirName,
**kwargs)
)
kwargs = config.get("ImagesGenerateHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs,
**kwargs}
hooks.append(ImagesGenerateHook(model=self,
dirName=self.dirName,
**kwargs
)
)
# kwargs = config.get("ImportanceSamplingHook", None)
# if kwargs:
# if not isinstance(kwargs, list):
# kwargs = [kwargs]
# for kw in kwargs:
# kws = {**self._plot_model_hooks_kwargs,
# **kw}
# hooks.append(ImportanceSamplingHook(model = self,
# dirName = self.dirName,
# tensors_to_average = [self.importance_sampling_node],
# datasets_keys = [TRAIN, VALIDATION], # don't change the order (Luigi)
# **kws
# )
# )
#
# kwargs = config.get("TwoDimPCALatentVariablesHook", None)
# if kwargs:
# kwargs = {**self._default_model_hooks_kwargs,
# **kwargs}
# hooks.append(TwoDimPCALatentVariablesHook(model = self,
# dirName = self.dirName,
# tensors = [self.z] + list(self._approximate_posterior_params),
# tensors_names = ['z',
# 'mu'],
# datasets_keys = [TRAIN, VALIDATION], # don't change the order (Luigi)
# **kwargs
# )
# )
#
# kwargs = config.get("PCALatentVariablesHook", None)
# if kwargs:
# kwargs = {**self._plot_model_hooks_kwargs,
# **kwargs}
# hooks.append(PCALatentVariablesHook(model = self,
# dirName = self.dirName,
# tensors = [self.z,
# self._approximate_posterior_params[0]],
# tensors_names = ['z',
# 'mu'],
# datasets_keys = [TRAIN, VALIDATION], # don't change the order (Luigi)
# **kwargs
# )
# )
#
# kwargs = config.get("VAELinearInterpolationHook", None)
# if kwargs:
# kwargs = {**self._default_model_hooks_kwargs,
# **kwargs}
# hooks.append(VAELinearInterpolationHook(model = self,
# dirName = self.dirName,
# **kwargs)
# )
#
#
# kwargs = config.get("LatentVarsClassificationHook", None)
# if kwargs:
# kwargs = {**self._plot_model_hooks_kwargs,
# **kwargs}
# hooks.append(LatentVarsClassificationHook(model = self,
# dirName = self.dirName,
# tensors = [self.z,
# self._approximate_posterior_params[0],
# tf.concat(list(self._approximate_posterior_params), axis=1)],
# tensors_names = ['z',
# 'mu',
# 'mu_cov'],
# datasets_keys = [TRAIN,VALIDATION], # don't change the order (Luigi)
# **kwargs)
# )
#
#
# # frechet inception distance
# kwargs = config.get("FrechetInceptionDistanceHook", None)
# if kwargs:
# for kw in kwargs:
# kws = {**self._default_model_hooks_kwargs,
# **kw}
# hooks.append(FrechetInceptionDistanceHook(model = self,
# dirName = self.dirName,
# **kws)
# )
#
# # latent traversals hook
# kwargs = config.get("LatentTraversalsHook", None)
# if kwargs:
# kwargs = {**self._default_model_hooks_kwargs,
# **kwargs}
# hooks.append(LatentTraversalsHook(model=self,
# dirName=self.dirName,
# **kwargs)
# )
#
# kwargs = config.get("LatentVarsGeometricClassificationHook", None)
# if kwargs:
# kwargs = {**self._default_model_hooks_kwargs,
# **kwargs}
# hooks.append(LatentVarsGeometricClassificationHook(model=self,
# dirName=self.dirName,
# tensors=[self.z,
# self._approximate_posterior_params[0],
# tf.concat(list(self._approximate_posterior_params),
# axis=1)],
# tensors_names=['z',
# 'mu',
# 'mu_cov'],
#
# datasets_keys=[TRAIN, VALIDATION, TEST],
# **kwargs)
# )
return hooks
def create_hooks(self, config):
hooks = super().create_hooks(config)
# logging hooks
log_tensors_to_average = self.loss_nodes_to_log
log_tensors_to_average_names = self.loss_nodes_to_log_names
log_tensors_to_average_plots = self.loss_nodes_to_log_filenames
hooks.append(
LoggingMeanTensorsHook(
model=self,
fileName="log",
dirName=self.dirName,
tensors_to_average=log_tensors_to_average,
tensors_to_average_names=log_tensors_to_average_names,
tensors_to_average_plots=log_tensors_to_average_plots,
time_reference=self._time_reference_str,
average_steps=self._n_steps_stats,
tensorboard_dir=self._tensorboard_dir,
trigger_summaries=config["save_summaries"],
#trigger_plot = True,
print_to_screen=True,
plot_offset=self._plot_offset,
train_loop_key=TRAIN_LOOP,
# if you want to remove some dataset from here, make support to specify from conf on which datasets to log, if in doubt ask me please. Riccardo
datasets_keys=[TRAIN, VALIDATION, TEST]))
kwargs = config.get("CorrelationHook", None)
if kwargs:
kwargs = {
**self._default_model_hooks_kwargs, 'datasets_keys':
[TRAIN, VALIDATION],
**kwargs
}
hooks.append(
CorrelationHook(model=self, dirName=self.dirName, **kwargs))
kwargs = config.get("MCDropoutHook", None)
if kwargs:
kwargs = {
**self._default_model_hooks_kwargs, 'datasets_keys':
[VALIDATION, TEST],
**kwargs
}
hooks.append(
MCDropoutHook(model=self, dirName=self.dirName, **kwargs))
kwargs = config.get("MCRegressionHook", None)
if kwargs:
kwargs = {
**self._default_model_hooks_kwargs, 'datasets_keys':
[VALIDATION, TEST],
**kwargs
}
hooks.append(
MCRegressionHook(model=self, dirName=self.dirName, **kwargs))
kwargs = config.get("MCClassificationHook", None)
if kwargs:
kwargs = {
**self._default_model_hooks_kwargs, 'datasets_keys':
[VALIDATION, TEST],
**kwargs
}
hooks.append(
MCClassificationHook(model=self,
dirName=self.dirName,
**kwargs))
# kwargs = config.get("MCDropoutHook_alpha", None)
# if kwargs:
# kwargs = {**self._default_model_hooks_kwargs,
# 'datasets_keys' : [TEST,VALIDATION],
# **kwargs}
#
# hooks.append(MCDropoutHook_alpha(model=self,
# dirName=self.dirName,
# **kwargs))
kwargs = config.get("WeightsHistogramHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs, **kwargs}
hooks.append(
WeightsHistogramHook(model=self,
dirName=self.dirName,
**kwargs))
'''
kwargs = config.get("HessianHook", None)
if kwargs:
kwargs = {**self._default_model_hooks_kwargs,
**kwargs}
hooks.append(HessianHook(model = self,
dirName = self.dirName,
#tensors = [self.z,
# self._gaussian_model_latent_mean],
#tensors_names = ['z',
# 'mu'],
datasets_keys = [TRAIN, VALIDATION], # don't change the order (Luigi)
**kwargs
)
)
'''
return hooks