class _SubsetParameters(ParameterGroup):
# Parameters governing sample distribution over subsets
header = string_attribute("Subset parameters")
description = string_attribute("Parameters for the different subsets")
# Add a parameter group 'Subset parameters'
auto_subset_fractions = configurable_boolean(
default_value=True,
description="Test",
header="Automatically determine subset proportions",
affects_outcome_of=ModelLifecycle.TRAINING,
)
__ui_rules = UIRules(
rules=[
Rule(
parameter="auto_subset_fractions",
value=False,
operator=Operator.EQUAL_TO,
)
],
action=Action.SHOW,
)
train_proportion = configurable_float(
default_value=0.75,
min_value=0.0,
max_value=1.0,
header="Training set proportion",
ui_rules=__ui_rules,
affects_outcome_of=ModelLifecycle.TRAINING,
)
validation_proportion = configurable_float(
default_value=0.1,
min_value=0.0,
max_value=1.0,
header="Validation set proportion",
ui_rules=__ui_rules,
affects_outcome_of=ModelLifecycle.TRAINING,
)
test_proportion = configurable_float(
default_value=0.15,
min_value=0.0,
max_value=1.0,
header="Test set proportion",
ui_rules=__ui_rules,
affects_outcome_of=ModelLifecycle.TRAINING,
)
class _SubgroupOne(ParameterGroup):
# Subgroup one of the nested group, with a couple of parameters
header = string_attribute("Parameter group one")
__ui_rules = UIRules(
rules=[
Rule(
parameter=[
"nested_parameter_group", "show_subgroup_one"
],
operator=Operator.EQUAL_TO,
value=False,
)
],
action=Action.HIDE,
)
bogus_parameter_one = configurable_float(
default_value=42,
ui_rules=__ui_rules,
header="Bogus parameter to test nested parameter groups",
)
bogus_parameter_two = configurable_float(
default_value=42,
ui_rules=__ui_rules,
header="Bogus parameter to test nested parameter groups",
)
class STFPMConfig(BaseAnomalyClassificationConfig):
"""
Configurable parameters for STFPM anomaly classification task.
"""
header = string_attribute("Configuration for STFPM")
description = header
@attrs
class ModelParameters(ParameterGroup):
"""
Parameter Group for training model
"""
header = string_attribute("Model Parameters")
description = header
@attrs
class EarlyStoppingParameters(ParameterGroup):
"""
Early stopping parameters
"""
header = string_attribute("Early Stopping Parameters")
description = header
metric = selectable(
default_value=EarlyStoppingMetrics.IMAGE_ROC_AUC,
header="Early Stopping Metric",
description="The metric used to determine if the model should stop training",
)
early_stopping = add_parameter_group(EarlyStoppingParameters)
model = add_parameter_group(ModelParameters)
class DatasetParameters(ParameterGroup):
"""
Parameters related to dataloader
"""
header = string_attribute("Dataset Parameters")
description = header
train_batch_size = configurable_integer(
default_value=32,
min_value=1,
max_value=512,
header="Batch size",
description=
"The number of training samples seen in each iteration of training. Increasing this value "
"improves training time and may make the training more stable. A larger batch size has higher "
"memory requirements.",
warning=
"Increasing this value may cause the system to use more memory than available, "
"potentially causing out of memory errors, please update with caution.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
num_workers = configurable_integer(
default_value=8,
min_value=0,
max_value=36,
header="Number of workers",
description=
"Increasing this value might improve training speed however it might cause out of memory "
"errors. If the number of workers is set to zero, data loading will happen in the main "
"training thread.",
)
class PadimConfig(BaseAnomalyClassificationConfig):
"""
Configurable parameters for PADIM anomaly classification task.
"""
header = string_attribute("Configuration for Padim")
description = header
class __LearningParameters(ParameterGroup):
header = string_attribute("Test Learning Parameters")
description = header
batch_size = configurable_integer(
default_value=5,
min_value=1,
max_value=512,
header="Test batch size",
description=
"The number of training samples seen in each iteration of training. Increasing this value "
"improves training time and may make the training more stable. A larger batch size has higher "
"memory requirements.",
warning=
"Increasing this value may cause the system to use more memory than available, "
"potentially causing out of memory errors, please update with caution.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
num_iters = configurable_integer(
default_value=1,
min_value=1,
max_value=100000,
header="Number of training iterations",
description=
"Increasing this value causes the results to be more robust but training time will be longer.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
learning_rate = configurable_float(
default_value=0.01,
min_value=1e-07,
max_value=1e-01,
header="Learning rate",
description=
"Increasing this value will speed up training convergence but might make it unstable.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
learning_rate_warmup_iters = configurable_integer(
default_value=100,
min_value=1,
max_value=10000,
header="Number of iterations for learning rate warmup",
description="Test learning rate warmup",
affects_outcome_of=ModelLifecycle.TRAINING,
)
num_workers = configurable_integer(
default_value=4,
min_value=2,
max_value=10,
header="num_workers test header",
description="num_workers test description",
affects_outcome_of=ModelLifecycle.NONE,
)
class _NestedParameterGroup(ParameterGroup):
# A nested group of parameters
# header is a required attribute, all parameter groups should define one.
header = string_attribute("Test group of parameter groups")
@attr.s
class _SubgroupOne(ParameterGroup):
# Subgroup one of the nested group, with a couple of parameters
header = string_attribute("Parameter group one")
__ui_rules = UIRules(
rules=[
Rule(
parameter=[
"nested_parameter_group", "show_subgroup_one"
],
operator=Operator.EQUAL_TO,
value=False,
)
],
action=Action.HIDE,
)
bogus_parameter_one = configurable_float(
default_value=42,
ui_rules=__ui_rules,
header="Bogus parameter to test nested parameter groups",
)
bogus_parameter_two = configurable_float(
default_value=42,
ui_rules=__ui_rules,
header="Bogus parameter to test nested parameter groups",
)
subgroup_one = add_parameter_group(_SubgroupOne)
show_subgroup_one = configurable_boolean(
default_value=True, header="Show the parameters in subgroup one?")
@attr.s
class _SubgroupTwo(ParameterGroup):
# Subgroup two of the nested group, with a couple of parameters
header = string_attribute("Parameter group two")
bogus_parameter_three = configurable_float(
default_value=42,
header="Bogus parameter to test nested parameter groups",
)
bogus_parameter_four = configurable_float(
default_value=42,
header="Bogus parameter to test nested parameter groups",
)
subgroup_two = add_parameter_group(_SubgroupTwo)
class _SubgroupTwo(ParameterGroup):
# Subgroup two of the nested group, with a couple of parameters
header = string_attribute("Parameter group two")
bogus_parameter_three = configurable_float(
default_value=42,
header="Bogus parameter to test nested parameter groups",
)
bogus_parameter_four = configurable_float(
default_value=42,
header="Bogus parameter to test nested parameter groups",
)
class EarlyStoppingParameters(ParameterGroup):
"""
Early stopping parameters
"""
header = string_attribute("Early Stopping Parameters")
description = header
metric = selectable(
default_value=EarlyStoppingMetrics.IMAGE_ROC_AUC,
header="Early Stopping Metric",
description="The metric used to determine if the model should stop training",
)
class __LearningParameters(ParameterGroup):
header = string_attribute("Learning Parameters")
description = header
batch_size = configurable_integer(
default_value=32,
min_value=1,
max_value=512,
header="Batch size",
description=
"The number of training samples seen in each iteration of training. Increasing this value "
"improves training time and may make the training more stable. A larger batch size has higher "
"memory requirements.",
warning=
"Increasing this value may cause the system to use more memory than available, "
"potentially causing out of memory errors, please update with caution.",
affects_outcome_of=ModelLifecycle.TRAINING)
max_num_epochs = configurable_integer(
default_value=200,
min_value=1,
max_value=1000,
header="Maximum number of training epochs",
description=
"Increasing this value causes the results to be more robust but training time "
"will be longer.",
affects_outcome_of=ModelLifecycle.TRAINING)
learning_rate = configurable_float(
default_value=0.01,
min_value=1e-07,
max_value=1e-01,
header="Learning rate",
description="Increasing this value will speed up training \
convergence but might make it unstable.",
affects_outcome_of=ModelLifecycle.TRAINING)
enable_lr_finder = configurable_boolean(
default_value=False,
header="Enable automatic learing rate estimation",
description=
"Learning rate parameter value will be ignored if enabled.",
affects_outcome_of=ModelLifecycle.TRAINING)
enable_early_stopping = configurable_boolean(
default_value=True,
header="Enable adaptive early stopping of the training",
description="Adaptive early exit from training when accuracy isn't \
changed or decreased for several epochs.",
affects_outcome_of=ModelLifecycle.TRAINING)
class __POTParameter(ParameterGroup):
header = string_attribute("POT Parameters")
description = header
stat_subset_size = configurable_integer(
header="Number of data samples",
description=
"Number of data samples used for post-training optimization",
default_value=300,
min_value=1,
max_value=maxsize)
preset = selectable(
default_value=POTQuantizationPreset.PERFORMANCE,
header="Preset",
description="Quantization preset that defines quantization scheme",
editable=False,
visible_in_ui=False)
class __Postprocessing(ParameterGroup):
header = string_attribute("Test Postprocessing")
description = header
result_based_confidence_threshold = configurable_boolean(
default_value=True,
header="Test Result based confidence threshold",
description="Test confidence threshold is derived from the results",
affects_outcome_of=ModelLifecycle.INFERENCE,
)
confidence_threshold = configurable_float(
default_value=0.25,
min_value=0,
max_value=1,
header="Test Confidence threshold",
description=
"This threshold only takes effect if the threshold is not set based on the result.--Only test",
affects_outcome_of=ModelLifecycle.INFERENCE,
)
class __NNCFOptimization(ParameterGroup):
header = string_attribute("Optimization by NNCF")
description = header
enable_quantization = configurable_boolean(
default_value=True,
header="Enable quantization algorithm",
description="Enable quantization algorithm",
affects_outcome_of=ModelLifecycle.TRAINING)
enable_pruning = configurable_boolean(
default_value=False,
header="Enable filter pruning algorithm",
description="Enable filter pruning algorithm",
affects_outcome_of=ModelLifecycle.TRAINING)
maximal_accuracy_degradation = configurable_float(
default_value=1.0,
min_value=0.0,
max_value=100.0,
header="Maximum accuracy degradation",
description="The maximal allowed accuracy metric drop",
affects_outcome_of=ModelLifecycle.TRAINING)
class ConfigExample(ConfigurableParameters):
header = string_attribute(
"Test configuration for an object detection task")
description = header
class __LearningParameters(ParameterGroup):
header = string_attribute("Test Learning Parameters")
description = header
batch_size = configurable_integer(
default_value=5,
min_value=1,
max_value=512,
header="Test batch size",
description=
"The number of training samples seen in each iteration of training. Increasing this value "
"improves training time and may make the training more stable. A larger batch size has higher "
"memory requirements.",
warning=
"Increasing this value may cause the system to use more memory than available, "
"potentially causing out of memory errors, please update with caution.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
num_iters = configurable_integer(
default_value=1,
min_value=1,
max_value=100000,
header="Number of training iterations",
description=
"Increasing this value causes the results to be more robust but training time will be longer.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
learning_rate = configurable_float(
default_value=0.01,
min_value=1e-07,
max_value=1e-01,
header="Learning rate",
description=
"Increasing this value will speed up training convergence but might make it unstable.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
learning_rate_warmup_iters = configurable_integer(
default_value=100,
min_value=1,
max_value=10000,
header="Number of iterations for learning rate warmup",
description="Test learning rate warmup",
affects_outcome_of=ModelLifecycle.TRAINING,
)
num_workers = configurable_integer(
default_value=4,
min_value=2,
max_value=10,
header="num_workers test header",
description="num_workers test description",
affects_outcome_of=ModelLifecycle.NONE,
)
class __Postprocessing(ParameterGroup):
header = string_attribute("Test Postprocessing")
description = header
result_based_confidence_threshold = configurable_boolean(
default_value=True,
header="Test Result based confidence threshold",
description="Test confidence threshold is derived from the results",
affects_outcome_of=ModelLifecycle.INFERENCE,
)
confidence_threshold = configurable_float(
default_value=0.25,
min_value=0,
max_value=1,
header="Test Confidence threshold",
description=
"This threshold only takes effect if the threshold is not set based on the result.--Only test",
affects_outcome_of=ModelLifecycle.INFERENCE,
)
learning_parameters = add_parameter_group(__LearningParameters)
postprocessing = add_parameter_group(__Postprocessing)
class BaseAnomalyClassificationConfig(ConfigurableParameters):
"""
Base OTE configurable parameters for anomaly classification task.
"""
header = string_attribute(
"Configuration for an anomaly classification task")
description = header
@attrs
class DatasetParameters(ParameterGroup):
"""
Parameters related to dataloader
"""
header = string_attribute("Dataset Parameters")
description = header
train_batch_size = configurable_integer(
default_value=32,
min_value=1,
max_value=512,
header="Batch size",
description=
"The number of training samples seen in each iteration of training. Increasing this value "
"improves training time and may make the training more stable. A larger batch size has higher "
"memory requirements.",
warning=
"Increasing this value may cause the system to use more memory than available, "
"potentially causing out of memory errors, please update with caution.",
affects_outcome_of=ModelLifecycle.TRAINING,
)
num_workers = configurable_integer(
default_value=8,
min_value=0,
max_value=36,
header="Number of workers",
description=
"Increasing this value might improve training speed however it might cause out of memory "
"errors. If the number of workers is set to zero, data loading will happen in the main "
"training thread.",
)
@attrs
class POTParameters(ParameterGroup):
"""
Training parameters for post-training optimization
"""
header = string_attribute("POT Parameters")
description = header
preset = selectable(
default_value=POTQuantizationPreset.PERFORMANCE,
header="Preset",
description="Quantization preset that defines quantization scheme",
)
stat_subset_size = configurable_integer(
header="Number of data samples",
description=
"Number of data samples used for post-training optimization",
default_value=300,
min_value=1,
max_value=maxsize,
)
dataset = add_parameter_group(DatasetParameters)
pot_parameters = add_parameter_group(POTParameters)
class OTEClassificationParameters(ConfigurableParameters):
header = string_attribute("Configuration for an image classification task")
description = header
@attrs
class __LearningParameters(ParameterGroup):
header = string_attribute("Learning Parameters")
description = header
batch_size = configurable_integer(
default_value=32,
min_value=1,
max_value=512,
header="Batch size",
description=
"The number of training samples seen in each iteration of training. Increasing this value "
"improves training time and may make the training more stable. A larger batch size has higher "
"memory requirements.",
warning=
"Increasing this value may cause the system to use more memory than available, "
"potentially causing out of memory errors, please update with caution.",
affects_outcome_of=ModelLifecycle.TRAINING)
max_num_epochs = configurable_integer(
default_value=200,
min_value=1,
max_value=1000,
header="Maximum number of training epochs",
description=
"Increasing this value causes the results to be more robust but training time "
"will be longer.",
affects_outcome_of=ModelLifecycle.TRAINING)
learning_rate = configurable_float(
default_value=0.01,
min_value=1e-07,
max_value=1e-01,
header="Learning rate",
description="Increasing this value will speed up training \
convergence but might make it unstable.",
affects_outcome_of=ModelLifecycle.TRAINING)
enable_lr_finder = configurable_boolean(
default_value=False,
header="Enable automatic learing rate estimation",
description=
"Learning rate parameter value will be ignored if enabled.",
affects_outcome_of=ModelLifecycle.TRAINING)
enable_early_stopping = configurable_boolean(
default_value=True,
header="Enable adaptive early stopping of the training",
description="Adaptive early exit from training when accuracy isn't \
changed or decreased for several epochs.",
affects_outcome_of=ModelLifecycle.TRAINING)
@attrs
class __NNCFOptimization(ParameterGroup):
header = string_attribute("Optimization by NNCF")
description = header
enable_quantization = configurable_boolean(
default_value=True,
header="Enable quantization algorithm",
description="Enable quantization algorithm",
affects_outcome_of=ModelLifecycle.TRAINING)
enable_pruning = configurable_boolean(
default_value=False,
header="Enable filter pruning algorithm",
description="Enable filter pruning algorithm",
affects_outcome_of=ModelLifecycle.TRAINING)
maximal_accuracy_degradation = configurable_float(
default_value=1.0,
min_value=0.0,
max_value=100.0,
header="Maximum accuracy degradation",
description="The maximal allowed accuracy metric drop",
affects_outcome_of=ModelLifecycle.TRAINING)
@attrs
class __POTParameter(ParameterGroup):
header = string_attribute("POT Parameters")
description = header
stat_subset_size = configurable_integer(
header="Number of data samples",
description=
"Number of data samples used for post-training optimization",
default_value=300,
min_value=1,
max_value=maxsize)
preset = selectable(
default_value=POTQuantizationPreset.PERFORMANCE,
header="Preset",
description="Quantization preset that defines quantization scheme",
editable=False,
visible_in_ui=False)
learning_parameters = add_parameter_group(__LearningParameters)
nncf_optimization = add_parameter_group(__NNCFOptimization)
pot_parameters = add_parameter_group(__POTParameter)
class DatasetManagerConfig(ConfigurableParameters):
"""Dummy configurable parameters class"""
# type: ignore
# This class is used for testing purposes only, so mypy should ignore it
# Component and header are required, description is optional.
header = string_attribute("Dataset Manager configuration -- TEST ONLY")
description = string_attribute(
"Configurable parameters for the DatasetManager -- TEST ONLY")
# Add some parameters
number_of_samples_for_auto_train = configurable_integer(
default_value=5,
min_value=1,
max_value=1000,
header="Samples required for new training round",
)
label_constraints = configurable_boolean(default_value=True,
header="Apply label constraints")
@attr.s
class _NestedParameterGroup(ParameterGroup):
# A nested group of parameters
# header is a required attribute, all parameter groups should define one.
header = string_attribute("Test group of parameter groups")
@attr.s
class _SubgroupOne(ParameterGroup):
# Subgroup one of the nested group, with a couple of parameters
header = string_attribute("Parameter group one")
__ui_rules = UIRules(
rules=[
Rule(
parameter=[
"nested_parameter_group", "show_subgroup_one"
],
operator=Operator.EQUAL_TO,
value=False,
)
],
action=Action.HIDE,
)
bogus_parameter_one = configurable_float(
default_value=42,
ui_rules=__ui_rules,
header="Bogus parameter to test nested parameter groups",
)
bogus_parameter_two = configurable_float(
default_value=42,
ui_rules=__ui_rules,
header="Bogus parameter to test nested parameter groups",
)
subgroup_one = add_parameter_group(_SubgroupOne)
show_subgroup_one = configurable_boolean(
default_value=True, header="Show the parameters in subgroup one?")
@attr.s
class _SubgroupTwo(ParameterGroup):
# Subgroup two of the nested group, with a couple of parameters
header = string_attribute("Parameter group two")
bogus_parameter_three = configurable_float(
default_value=42,
header="Bogus parameter to test nested parameter groups",
)
bogus_parameter_four = configurable_float(
default_value=42,
header="Bogus parameter to test nested parameter groups",
)
subgroup_two = add_parameter_group(_SubgroupTwo)
@attr.s
class _SubsetParameters(ParameterGroup):
# Parameters governing sample distribution over subsets
header = string_attribute("Subset parameters")
description = string_attribute("Parameters for the different subsets")
# Add a parameter group 'Subset parameters'
auto_subset_fractions = configurable_boolean(
default_value=True,
description="Test",
header="Automatically determine subset proportions",
affects_outcome_of=ModelLifecycle.TRAINING,
)
__ui_rules = UIRules(
rules=[
Rule(
parameter="auto_subset_fractions",
value=False,
operator=Operator.EQUAL_TO,
)
],
action=Action.SHOW,
)
train_proportion = configurable_float(
default_value=0.75,
min_value=0.0,
max_value=1.0,
header="Training set proportion",
ui_rules=__ui_rules,
affects_outcome_of=ModelLifecycle.TRAINING,
)
validation_proportion = configurable_float(
default_value=0.1,
min_value=0.0,
max_value=1.0,
header="Validation set proportion",
ui_rules=__ui_rules,
affects_outcome_of=ModelLifecycle.TRAINING,
)
test_proportion = configurable_float(
default_value=0.15,
min_value=0.0,
max_value=1.0,
header="Test set proportion",
ui_rules=__ui_rules,
affects_outcome_of=ModelLifecycle.TRAINING,
)
# Add a selectable and float selectable parameter
dummy_float_selectable = float_selectable(options=[1.0, 2.0, 3.0, 4.0],
default_value=2.0,
header="Test float selectable")
dummy_selectable = selectable(
default_value=SomeEnumSelectable.BOGUS_NAME,
header="Test",
affects_outcome_of=ModelLifecycle.INFERENCE,
)
# Finally, add the nested parameter group and subset parameter groups to the config
# NOTE! group initialization should use a factory to avoid passing mutable default arguments. This is why the
# add_parameter_group function is needed.
nested_parameter_group = add_parameter_group(_NestedParameterGroup)
subset_parameters = add_parameter_group(_SubsetParameters)
