def configure_ermlp_training_pipeline(model_name: str):
"""Configure ERMLP from pipeline.
:param model_name: name of the model
:rtype: OrderedDict
:return: configuration dictionary
"""
config = get_config_dict(model_name)
# Step 1: Query embedding dimension
print_training_embedding_dimension_message()
print_embedding_dimension_info_message()
embedding_dimension = select_integer_value(
print_msg=EMBEDDING_DIMENSION_PRINT_MSG,
prompt_msg=EMBEDDING_DIMENSION_PROMPT_MSG,
error_msg=EMBEDDING_DIMENSION_ERROR_MSG)
config[EMBEDDING_DIM] = embedding_dimension
print_section_divider()
# Step 2: Query margin loss
print_training_margin_loss_message()
magin_loss = select_float_value(print_msg=MARGIN_LOSS_PRINT_MSG,
prompt_msg=MARGIN_LOSS_PROMPT_MSG,
error_msg=MARGIN_LOSS_ERROR_MSG)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query learning rate
print_learning_rate_message()
learning_rate = select_float_value(print_msg=LEARNING_RATE_PRINT_MSG,
prompt_msg=LEARNING_RATE_PROMPT_MSG,
error_msg=LEARNING_RATE_ERROR_MSG)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 4: Query batch size
print_batch_size_message()
batch_size = select_integer_value(print_msg=BATCH_SIZE_PRINT_MSG,
prompt_msg=BATCH_SIZE_PROMPT_MSG,
error_msg=BATCH_SIZE_ERROR_MSG)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 5: Query number of epochs
print_number_epochs_message()
number_epochs = select_integer_value(print_msg=EPOCH_PRINT_MSG,
prompt_msg=EPOCH_PROMPT_MSG,
error_msg=EPOCH_ERROR_MSG)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def prompt_evaluation_parameters(config: Dict) -> None:
"""Prompt the user for evaluation parameters absed on the execution mode."""
if config[EXECUTION_MODE] == TRAINING_MODE:
# Step 1: Ask whether to evaluate the model
print_ask_for_evlauation_message()
is_evaluation_mode = click.confirm(
'Do you want to evaluate your model?')
print_section_divider()
else:
is_evaluation_mode = True
# Step 2: Specify test set, if is_evaluation_mode==True
if is_evaluation_mode:
print_test_set_message()
provide_test_set = click.confirm('Do you provide a test set yourself?')
print_section_divider()
if provide_test_set:
test_set_path = get_input_path(TEST_FILE_PROMPT_MSG)
config[TEST_SET_PATH] = test_set_path
else:
print_test_ratio_message()
test_set_ratio = select_ratio_for_test_set()
config[TEST_SET_RATIO] = test_set_ratio
print_section_divider()
# Ask whether to use filtered negative triples
print_filter_negative_triples_message()
filter_negative_triples = ask_for_filtering_of_negatives()
config[FILTER_NEG_TRIPLES] = filter_negative_triples
print_section_divider()
def configure_ermlp_hpo_pipeline(model_name: str):
config = get_config_dict(model_name)
# Step 1: Query embedding dimensions
print_hpo_embedding_dimensions_message()
embedding_dimensions = select_positive_integer_values(
EMBEDDING_DIMENSIONS_PRINT_MSG, EMBEDDING_DIMENSIONS_PROMPT_MSG,
EMBEDDING_DIMENSIONS_ERROR_MSG)
config[EMBEDDING_DIM] = embedding_dimensions
print_section_divider()
# Step 2: Query margin loss
print_hpo_margin_losses_message()
magin_loss = select_float_values(print_msg=MARGIN_LOSSES_PRINT_MSG,
prompt_msg=MARGIN_LOSSES_PROMPT_MSG,
error_msg=MARGIN_LOSSES_ERROR_MSG)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query learning rate
print_hpo_learning_rates_message()
learning_rate = select_float_values(print_msg=LEARNING_RATES_PRINT_MSG,
prompt_msg=LEARNING_RATES_PROMPT_MSG,
error_msg=LEARNING_RATES_ERROR_MSG)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 4: Query batch size
print_hpo_batch_sizes_message()
batch_sizes = select_positive_integer_values(
print_msg=BATCH_SIZES_PRINT_MSG,
prompt_msg=BATCH_SIZES_PROMPT_MSG,
error_msg=BATCH_SIZES_ERROR_MSG)
config[BATCH_SIZE] = batch_sizes
print_section_divider()
# Step 5: Query number of epochs
print_hpo_epochs_message()
number_epochs = select_positive_integer_values(
print_msg=EPOCHS_PRINT_MSG,
prompt_msg=EPOCHS_PROMPT_MSG,
error_msg=EPOCHS_ERROR_MSG)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def prompt_execution_parameters(config: Dict, model_name: str) -> None:
"""Prompt the user for execution mode parameters."""
pykeen_exec_mode = config[EXECUTION_MODE]
if pykeen_exec_mode == TRAINING_MODE:
config.update(_configure_training_pipeline(model_name))
elif pykeen_exec_mode == HPO_MODE:
config.update(_configure_hpo_pipeline(model_name))
# Query number of HPO iterations
hpo_iter = select_integer_value(
print_msg=HPO_ITERS_PRINT_MSG,
prompt_msg=HPO_ITERS_PROMPT_MSG,
error_msg=HPO_ITERS_ERROR_MSG,
)
config[NUM_OF_HPO_ITERS] = hpo_iter
print_section_divider()
def prompt_biokeen_config(*,
connection: str,
rebuild: bool,
do_prompt_bio2bel: Optional[bool] = None) -> Dict:
"""Configure experiments."""
config = OrderedDict()
# Step 1: Welcome + Intro
print_welcome_message()
print_section_divider()
print_intro()
print_section_divider()
# Step 2: Ask for data source
if do_prompt_bio2bel is None:
do_prompt_bio2bel = click.confirm(
'Do you want to use one of the databases provided by BioKEEN?',
default=True)
print_section_divider()
do_prompt_training = True
if do_prompt_bio2bel:
do_prompt_training = False
config[TRAINING_SET_PATH] = []
for name in select_bio2bel_repository():
try:
path = install_bio2bel_module(name=name,
connection=connection,
rebuild=rebuild)
except Exception:
click.secho(f'failed: {name}', fg='red')
else:
if os.path.exists(path):
config[TRAINING_SET_PATH].append(f'bio2bel:{name}')
else:
click.secho(f'failed: {name}: {path}', fg='red')
# TODO replace this with less safe code that assumes everything installs no problemo
"""
config[TRAINING_SET_PATH] = [
f'bio2bel:{name}'
for name in select_bio2bel_repository()
]
"""
print_section_divider()
return prompt_config(
config=config,
show_welcome=False,
do_prompt_training=do_prompt_training,
)
def configure_um_training_pipeline(model_name: str):
config = OrderedDict()
config[KG_EMBEDDING_MODEL_NAME] = model_name
# Step 1: Query embedding dimension
print_training_embedding_dimension_message()
print_embedding_dimension_info_message()
embedding_dimension = select_integer_value(print_msg=EMBEDDING_DIMENSION_PRINT_MSG,
prompt_msg=EMBEDDING_DIMENSION_PROMPT_MSG,
error_msg=EMBEDDING_DIMENSION_ERROR_MSG)
config[EMBEDDING_DIM] = embedding_dimension
print_section_divider()
# Step 2: Query margin loss
print_training_margin_loss_message()
magin_loss = select_float_value(print_msg=MARGIN_LOSS_PRINT_MSG,
prompt_msg=MARGIN_LOSS_PROMPT_MSG,
error_msg=MARGIN_LOSS_ERROR_MSG)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query L_p norm as scoring function
print_scoring_fct_message()
scoring_fct_norm = select_integer_value(print_msg=NORM_SCORING_FUNCTION_PRINT_MSG,
prompt_msg=NORM_SCORING_FUNCTION_PROMPT_MSG,
error_msg=NORM_SCORING_FUNCTION_ERROR_MSG)
config[SCORING_FUNCTION_NORM] = scoring_fct_norm
print_section_divider()
# Step 4: Query L_p norm for normalizing the entities
print_entity_normalization_message()
entity_normalization_norm = select_integer_value(print_msg=ENTITIES_NORMALIZATION_PRINT_MSG,
prompt_msg=ENTITIES_NORMALIZATION_PROMPT_MSG,
error_msg=ENTITIES_NORMALIZATION_ERROR_MSG)
config[NORM_FOR_NORMALIZATION_OF_ENTITIES] = entity_normalization_norm
print_section_divider()
# Step 5: Query learning rate
print_learning_rate_message()
learning_rate = select_float_value(print_msg=LEARNING_RATE_PRINT_MSG,
prompt_msg=LEARNING_RATE_PROMPT_MSG,
error_msg=LEARNING_RATE_ERROR_MSG)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 6: Query batch size
print_batch_size_message()
batch_size = select_integer_value(print_msg=BATCH_SIZE_PRINT_MSG,
prompt_msg=BATCH_SIZE_PROMPT_MSG,
error_msg=BATCH_SIZE_ERROR_MSG)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 7: Query number of epochs
print_number_epochs_message()
number_epochs = select_integer_value(print_msg=EPOCH_PRINT_MSG,
prompt_msg=EPOCH_PROMPT_MSG,
error_msg=EPOCH_ERROR_MSG)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def configure_um_hpo_pipeline(model_name):
config = get_config_dict(model_name)
# Step 1: Query embedding dimensions
print_hpo_embedding_dimensions_message()
embedding_dimensions = select_positive_integer_values(EMBEDDING_DIMENSIONS_PRINT_MSG,
EMBEDDING_DIMENSIONS_PROMPT_MSG,
EMBEDDING_DIMENSIONS_ERROR_MSG)
config[EMBEDDING_DIM] = embedding_dimensions
print_section_divider()
# Step 2: Query margin loss
print_hpo_margin_losses_message()
magin_loss = select_float_values(
print_msg=MARGIN_LOSSES_PRINT_MSG,
prompt_msg=MARGIN_LOSSES_PROMPT_MSG,
error_msg=MARGIN_LOSSES_ERROR_MSG)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query L_p norms to use as scoring function
print_hpo_scoring_fcts_message()
scoring_fct_norm = select_positive_integer_values(
print_msg=NORMS_SCORING_FUNCTION_PRINT_MSG,
prompt_msg=NORMS_SCORING_FUNCTION_PROMPT_MSG,
error_msg=NORMS_SCORING_FUNCTION_ERROR_MSG
)
config[SCORING_FUNCTION_NORM] = scoring_fct_norm
print_section_divider()
# Step 4: Query L_p norms for normalizing the entities
print_hpo_entity_normalization_norms_message()
entity_normalization_norm = select_positive_integer_values(
print_msg=NORMS_FOR_NORMALIZATION_OF_ENTITIES_PRINT_MSG,
prompt_msg=NORMS_FOR_NORMALIZATION_OF_ENTITIES_PROMPT_MSG,
error_msg=NORMS_FOR_NORMALIZATION_OF_ENTITIES_ERROR_MSG
)
config[NORM_FOR_NORMALIZATION_OF_ENTITIES] = entity_normalization_norm
print_section_divider()
# Step 5: Query learning rate
print_hpo_learning_rates_message()
learning_rate = select_float_values(
print_msg=LEARNING_RATES_PRINT_MSG,
prompt_msg=LEARNING_RATES_PROMPT_MSG,
error_msg=LEARNING_RATES_ERROR_MSG)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 6: Query batch size
print_hpo_batch_sizes_message()
batch_size = select_positive_integer_values(
print_msg=BATCH_SIZES_PRINT_MSG,
prompt_msg=BATCH_SIZES_PROMPT_MSG,
error_msg=BATCH_SIZES_ERROR_MSG)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 7: Query number of epochs
print_hpo_epochs_message()
number_epochs = select_positive_integer_values(
print_msg=EPOCHS_PRINT_MSG,
prompt_msg=EPOCHS_PROMPT_MSG,
error_msg=EPOCHS_ERROR_MSG)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def configure_trans_d_training_pipeline(model_name: str):
"""Configure Trans D from pipeline.
:param model_name: name of the model
:return: configuration dictionary
"""
config = get_config_dict(model_name)
# Step 1: Query embedding dimension for entities
print_entities_embedding_dimension_message()
embedding_dimension = select_integer_value(
print_msg=ENTITIES_EMBEDDING_DIMENSION_PRINT_MSG,
prompt_msg=ENTITIES_EMBEDDING_DIMENSION_PROMPT_MSG,
error_msg=ENTITIES_EMBEDDING_DIMENSION_ERROR_MSG,
)
config[EMBEDDING_DIM] = embedding_dimension
print_section_divider()
# Step 2: Query embedding dimension for relations
print_relations_embedding_dimension_message()
relation_embedding_dimension = select_integer_value(
print_msg=RELATION_EMBEDDING_DIMENSION_PRINT_MSG,
prompt_msg=RELATION_EMBEDDING_DIMENSION_PROMPT_MSG,
error_msg=RELATION_EMBEDDING_DIMENSION_ERROR_MSG,
)
config[RELATION_EMBEDDING_DIM] = relation_embedding_dimension
print_section_divider()
# Step 2: Query margin loss
print_training_margin_loss_message()
magin_loss = select_float_value(print_msg=MARGIN_LOSS_PRINT_MSG,
prompt_msg=MARGIN_LOSS_PROMPT_MSG,
error_msg=MARGIN_LOSS_ERROR_MSG)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query L_p norm as scoring function
print_scoring_fct_message()
scoring_fct_norm = select_integer_value(
print_msg=NORM_SCORING_FUNCTION_PRINT_MSG,
prompt_msg=NORM_SCORING_FUNCTION_PROMPT_MSG,
error_msg=NORM_SCORING_FUNCTION_ERROR_MSG)
config[SCORING_FUNCTION_NORM] = scoring_fct_norm
print_section_divider()
# Step 5: Query learning rate
print_learning_rate_message()
learning_rate = select_float_value(print_msg=LEARNING_RATE_PRINT_MSG,
prompt_msg=LEARNING_RATE_PROMPT_MSG,
error_msg=LEARNING_RATE_ERROR_MSG)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 6: Query batch size
print_batch_size_message()
batch_size = select_integer_value(print_msg=BATCH_SIZE_PRINT_MSG,
prompt_msg=BATCH_SIZE_PROMPT_MSG,
error_msg=BATCH_SIZE_ERROR_MSG)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 7: Query number of epochs
print_number_epochs_message()
number_epochs = select_integer_value(print_msg=EPOCH_PRINT_MSG,
prompt_msg=EPOCH_PROMPT_MSG,
error_msg=EPOCH_ERROR_MSG)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def prompt_config(*,
config: Optional[Dict] = None,
show_welcome: bool = True,
do_prompt_training: bool = True) -> Dict:
"""Prompt the user for the run configuration."""
if config is None:
config = OrderedDict()
# Step 1: Welcome + Intro
if show_welcome:
print_welcome_message()
print_section_divider()
print_intro()
print_section_divider()
# Step 2: Ask for training file
if do_prompt_training:
prompt_training_file(config)
print_section_divider()
# Step 3: Ask for execution mode
prompt_execution_mode(config)
print_section_divider()
# Step 4: Ask for model
model_name = prompt_embedding_model()
print_section_divider()
# Step 5: Query parameters depending on the selected execution mode
prompt_execution_parameters(config, model_name=model_name)
print_section_divider()
# Step 5.5: Prompt for evaluation parameters depending on the selected execution mode
prompt_evaluation_parameters(config)
# Step 6: Please select a random seed
prompt_random_seed(config)
print_section_divider()
# Step 7: Query device to train on
prompt_device(config)
print_section_divider()
# Step 8: Define output directory
config[OUTPUT_DIREC] = query_output_directory()
print_section_divider()
config_path = os.path.join(config[OUTPUT_DIREC], 'configuration.json')
with open(config_path, 'w') as file:
json.dump(config, file, indent=2)
return config
def prompt_embedding_model() -> str:
"""Prompt the user to select an embedding model."""
print_model_selection_message()
model_name = select_embedding_model()
print_section_divider()
return model_name
def configure_trans_h_training_pipeline(model_name: str) -> Dict:
"""Prompt the user to configure Trans H from pipeline."""
config = get_config_dict(model_name)
# Step 1: Query embedding dimension
print_training_embedding_dimension_message()
print_embedding_dimension_info_message()
embedding_dimension = select_integer_value(
print_msg=EMBEDDING_DIMENSION_PRINT_MSG,
prompt_msg=EMBEDDING_DIMENSION_PROMPT_MSG,
error_msg=EMBEDDING_DIMENSION_ERROR_MSG,
)
config[EMBEDDING_DIM] = embedding_dimension
print_section_divider()
# Step 2: Query margin loss
print_training_margin_loss_message()
magin_loss = select_float_value(
print_msg=MARGIN_LOSS_PRINT_MSG,
prompt_msg=MARGIN_LOSS_PROMPT_MSG,
error_msg=MARGIN_LOSS_ERROR_MSG,
)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query L_p norm as scoring function
print_scoring_fct_message()
scoring_fct_norm = select_integer_value(
print_msg=NORM_SCORING_FUNCTION_PRINT_MSG,
prompt_msg=NORM_SCORING_FUNCTION_PROMPT_MSG,
error_msg=NORM_SCORING_FUNCTION_ERROR_MSG
)
config[SCORING_FUNCTION_NORM] = scoring_fct_norm
print_section_divider()
# Step 4: Query weight for the soft constraints
print_trans_h_soft_constraints_weight_message()
soft_constraints_weight = select_float_value(
print_msg=WEIGHTS_SOFT_CONSTRAINT_TRANS_H_PRINT_MSG,
prompt_msg=WEIGHTS_SOFT_CONSTRAINT_TRANS_H_PROMPT_MSG,
error_msg=WEIGHTS_SOFT_CONSTRAINT_TRANS_H_ERROR_MSG,
)
config[WEIGHT_SOFT_CONSTRAINT_TRANS_H] = soft_constraints_weight
print_section_divider()
# Step 5: Query learning rate
print_learning_rate_message()
learning_rate = select_float_value(
print_msg=LEARNING_RATE_PRINT_MSG,
prompt_msg=LEARNING_RATE_PROMPT_MSG,
error_msg=LEARNING_RATE_ERROR_MSG,
)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 6: Query batch size
print_batch_size_message()
batch_size = select_integer_value(
print_msg=BATCH_SIZE_PRINT_MSG,
prompt_msg=BATCH_SIZE_PROMPT_MSG,
error_msg=BATCH_SIZE_ERROR_MSG,
)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 7: Query number of epochs
print_number_epochs_message()
number_epochs = select_integer_value(
print_msg=EPOCH_PRINT_MSG,
prompt_msg=EPOCH_PROMPT_MSG,
error_msg=EPOCH_ERROR_MSG,
)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def configure_trans_h_hpo_pipeline(model_name):
config = get_config_dict(model_name)
# Step 1: Query embedding dimensions
print_hpo_embedding_dimensions_message()
embedding_dimensions = select_positive_integer_values(
EMBEDDING_DIMENSIONS_PRINT_MSG,
EMBEDDING_DIMENSIONS_PROMPT_MSG,
EMBEDDING_DIMENSIONS_ERROR_MSG,
)
config[EMBEDDING_DIM] = embedding_dimensions
print_section_divider()
# Step 2: Query margin loss
print_hpo_margin_losses_message()
magin_loss = select_float_values(
print_msg=MARGIN_LOSSES_PRINT_MSG,
prompt_msg=MARGIN_LOSSES_PROMPT_MSG,
error_msg=MARGIN_LOSSES_ERROR_MSG,
)
config[MARGIN_LOSS] = magin_loss
print_section_divider()
# Step 3: Query L_p norms to use as scoring function
print_hpo_scoring_fcts_message()
scoring_fct_norm = select_positive_integer_values(
print_msg=NORMS_SCORING_FUNCTION_PRINT_MSG,
prompt_msg=NORMS_SCORING_FUNCTION_PROMPT_MSG,
error_msg=NORMS_SCORING_FUNCTION_ERROR_MSG,
)
config[SCORING_FUNCTION_NORM] = scoring_fct_norm
print_section_divider()
# Step 4: Query weight for the soft constraints
print_hpo_trans_h_soft_constraints_weights_message()
soft_constraints_weight = select_float_values(
print_msg=WEIGHTS_SOFT_CONSTRAINT_TRANS_H_PRINT_MSG,
prompt_msg=WEIGHTS_SOFT_CONSTRAINT_TRANS_H_PROMPT_MSG,
error_msg=WEIGHTS_SOFT_CONSTRAINT_TRANS_H_ERROR_MSG,
)
config[WEIGHT_SOFT_CONSTRAINT_TRANS_H] = soft_constraints_weight
print_section_divider()
# Step 5: Query learning rate
print_hpo_learning_rates_message()
learning_rate = select_float_values(
print_msg=LEARNING_RATES_PRINT_MSG,
prompt_msg=LEARNING_RATES_PROMPT_MSG,
error_msg=LEARNING_RATES_ERROR_MSG,
)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 6: Query batch size
print_hpo_batch_sizes_message()
batch_sizes = select_positive_integer_values(
print_msg=BATCH_SIZES_PRINT_MSG,
prompt_msg=BATCH_SIZES_PROMPT_MSG,
error_msg=BATCH_SIZES_ERROR_MSG,
)
config[BATCH_SIZE] = batch_sizes
print_section_divider()
# Step 7: Query number of epochs
print_hpo_epochs_message()
number_epochs = select_positive_integer_values(
print_msg=EPOCHS_PRINT_MSG,
prompt_msg=EPOCHS_PROMPT_MSG,
error_msg=EPOCHS_ERROR_MSG,
)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def configure_conv_e_hpo_pipeline(model_name):
"""Configure ConvE.
:param str model_name: name of the model
:rtype: OrderedDict
:return: configuration dictionary
"""
config = get_config_dict(model_name)
# Step 1: Query embedding dimension
print_hpo_embedding_dimensions_message()
embedding_dimensions = select_positive_integer_values(
EMBEDDING_DIMENSIONS_PRINT_MSG,
EMBEDDING_DIMENSIONS_PROMPT_MSG,
EMBEDDING_DIMENSIONS_ERROR_MSG,
)
config[EMBEDDING_DIM] = embedding_dimensions
print_section_divider()
# Step 2: Query height and width
print_hpo_conv_e_width_height_message()
heights, widths = select_heights_and_widths(embedding_dimensions)
config[CONV_E_HEIGHT] = heights
config[CONV_E_WIDTH] = widths
print_section_divider()
# Step 3: Query number of input channels
print_conv_input_channels_message()
num_input_channels = select_positive_integer_values(
CONV_E_HPO_INPUT_CHANNELS_PRINT_MSG,
CONV_E_HPO_INPUT_CHANNELS_PROMPT_MSG,
CONV_E_HPO_INPUT_CHANNELS_ERROR_MSG,
)
config[CONV_E_INPUT_CHANNELS] = num_input_channels
print_section_divider()
# Step 4: Query number of output channels
print_conv_e_output_channels_message()
num_output_channels = select_positive_integer_values(
CONV_E_HPO_OUT_CHANNELS_PRINT_MSG,
CONV_E_HPO_OUT_CHANNELS_PROMPT_MSG,
CONV_E_HPO_OUT_CHANNELS_ERROR_MSG,
)
config[CONV_E_OUTPUT_CHANNELS] = num_output_channels
print_section_divider()
# Step 4: Query kernel height
print_conv_e_hpo_kernel_height_message()
kernel_heights = select_kernel_sizes(
depending_params=heights,
print_msg=CONV_E_HPO_KERNEL_HEIGHTS_PRINT_MSG,
prompt_msg=CONV_E_HPO_KERNEL_HEIGHTS_PROMPT_MSG,
error_msg=CONV_E_HPO_KERNEL_HEIGHTS_ERROR_MSG,
)
config[CONV_E_KERNEL_HEIGHT] = kernel_heights
print_section_divider()
# Step 5: Query kernel width
print_conv_e_hpo_kernel_width_message()
kernel_widths = select_kernel_sizes(
depending_params=widths,
print_msg=CONV_E_KERNEL_WIDTH_PRINT_MSG,
prompt_msg=CONV_E_KERNEL_WIDTH_PROMPT_MSG,
error_msg=CONV_E_KERNEL_WIDTH_ERROR_MSG,
)
config[CONV_E_KERNEL_WIDTH] = kernel_widths
print_section_divider()
# Step 6: Query dropout for input layer
print_hpo_input_dropout_message()
input_dropout = select_float_values(
CONV_E_HPO_INPUT_DROPOUTS_PRINT_MSG,
CONV_E_HPO_INPUT_DROPOUTS_PROMPT_MSG,
CONV_E_HPO_INPUT_DROPOUTS_ERROR_MSG,
)
config[CONV_E_INPUT_DROPOUT] = input_dropout
print_section_divider()
# Step 7: Query dropout for output layer
print_hpo_output_dropout_message()
output_dropouts = select_zero_one_range_float_values(
print_msg=CONV_E_HPO_OUTPUT_DROPOUT_PRINT_MSG,
prompt_msg=CONV_E_HPO_OUTPUT_DROPOUT_PROMPT_MSG,
error_msg=CONV_E_HPO_OUTPUT_DROPOUT_ERROR_MSG,
)
config[CONV_E_OUTPUT_DROPOUT] = output_dropouts
print_section_divider()
# Step 8: Query feature map dropout for output layer
print_hpo_feature_maps_dropouts_message()
feature_map_dropouts = select_zero_one_range_float_values(
print_msg=CONV_E_HPO_FEATURE_MAP_DROPOUT_PRINT_MSG,
prompt_msg=CONV_E_HPO_FEATURE_MAP_DROPOUT_PROMPT_MSG,
error_msg=CONV_E_HPO_FEATURE_MAP_DROPOUT_ERROR_MSG,
)
config[CONV_E_FEATURE_MAP_DROPOUT] = feature_map_dropouts
print_section_divider()
# Step 9: Query learning rate
print_hpo_learning_rates_message()
learning_rates = select_float_values(
print_msg=LEARNING_RATES_PRINT_MSG,
prompt_msg=LEARNING_RATES_PROMPT_MSG,
error_msg=LEARNING_RATES_ERROR_MSG,
)
config[LEARNING_RATE] = learning_rates
print_section_divider()
# Step 10: Query batch size
print_hpo_batch_sizes_message()
batch_size = select_positive_integer_values(
print_msg=BATCH_SIZES_PRINT_MSG,
prompt_msg=BATCH_SIZES_PROMPT_MSG,
error_msg=BATCH_SIZES_ERROR_MSG,
)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 11: Query number of epochs
print_hpo_epochs_message()
number_epochs = select_positive_integer_values(
print_msg=EPOCHS_PRINT_MSG,
prompt_msg=EPOCHS_PROMPT_MSG,
error_msg=EPOCHS_ERROR_MSG,
)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
def configure_conv_e_training_pipeline(model_name: str):
"""Configure ConvE.
:param str model_name: name of the model
:return: configuration dictionary
"""
config = get_config_dict(model_name)
# Step 1: Query embedding dimension
print_training_embedding_dimension_message()
print_embedding_dimension_info_message()
embedding_dimension = select_integer_value(
print_msg=EMBEDDING_DIMENSION_PRINT_MSG,
prompt_msg=EMBEDDING_DIMENSION_PROMPT_MSG,
error_msg=EMBEDDING_DIMENSION_ERROR_MSG,
)
config[EMBEDDING_DIM] = embedding_dimension
print_section_divider()
# Step 2: Query height and width
print_conv_e_width_height_message()
height, width = query_height_and_width_for_conv_e(embedding_dimension)
config[CONV_E_HEIGHT] = height
config[CONV_E_WIDTH] = width
print_section_divider()
# Step 3: Query number of input channels
print_conv_input_channels_message()
num_input_channels = select_integer_value(
CONV_E_INPUT_CHANNEL_PRINT_MSG,
CONV_E_INPUT_CHANNEL_PROMPT_MSG,
CONV_E_INPUT_CHANNEL_ERROR_MSG,
)
config[CONV_E_INPUT_CHANNELS] = num_input_channels
print_section_divider()
# Step 4: Query number of output channels
print_conv_e_output_channels_message()
num_output_channels = select_integer_value(
CONV_E_OUT_CHANNEL_PRINT_MSG,
CONV_E_OUT_CHANNEL_PROMPT_MSG,
CONV_E_OUT_CHANNEL_ERROR_MSG,
)
config[CONV_E_OUTPUT_CHANNELS] = num_output_channels
print_section_divider()
# Step 4: Query kernel height
print_conv_kernel_height_message()
kernel_height = query_kernel_param(
depending_param=height,
print_msg=CONV_E_KERNEL_HEIGHT_PRINT_MSG,
prompt_msg=CONV_E_KERNEL_HEIGHT_PROMPT_MSG,
error_msg=CONV_E_KERNEL_HEIGHT_ERROR_MSG,
)
config[CONV_E_KERNEL_HEIGHT] = kernel_height
print_section_divider()
# Step 5: Query kernel width
print_conv_kernel_width_message()
kernel_width = query_kernel_param(
depending_param=width,
print_msg=CONV_E_KERNEL_WIDTH_PRINT_MSG,
prompt_msg=CONV_E_KERNEL_WIDTH_PROMPT_MSG,
error_msg=CONV_E_KERNEL_WIDTH_ERROR_MSG,
)
config[CONV_E_KERNEL_WIDTH] = kernel_width
print_section_divider()
# Step 6: Query dropout for input layer
print_hpo_input_dropout_message()
input_dropout = select_zero_one_float_value(
print_msg=CONV_E_INPUT_DROPOUT_PRINT_MSG,
prompt_msg=CONV_E_INPUT_DROPOUT_PROMPT_MSG,
error_msg=CONV_E_INPUT_DROPOUT_ERROR_MSG,
)
config[CONV_E_INPUT_DROPOUT] = input_dropout
print_section_divider()
# Step 7: Query dropout for output layer
print_output_dropout_message()
output_dropout = select_zero_one_float_value(
print_msg=CONV_E_OUTPUT_DROPOUT_PRINT_MSG,
prompt_msg=CONV_E_OUTPUT_DROPOUT_PROMPT_MSG,
error_msg=CONV_E_OUTPUT_DROPOUT_ERROR_MSG,
)
config[CONV_E_OUTPUT_DROPOUT] = output_dropout
print_section_divider()
# Step 8: Query feature map dropout for output layer
print_feature_map_dropout_message()
feature_map_dropout = select_zero_one_float_value(
print_msg=CONV_E_FEATURE_MAP_DROPOUT_PRINT_MSG,
prompt_msg=CONV_E__FEATURE_MAP_DROPOUT_PROMPT_MSG,
error_msg=CONV_E_FEATURE_MAP_DROPOUT_ERROR_MSG,
)
config[CONV_E_FEATURE_MAP_DROPOUT] = feature_map_dropout
print_section_divider()
# Step 5: Query learning rate
print_learning_rate_message()
learning_rate = select_float_value(print_msg=LEARNING_RATE_PRINT_MSG,
prompt_msg=LEARNING_RATE_PROMPT_MSG,
error_msg=LEARNING_RATE_ERROR_MSG)
config[LEARNING_RATE] = learning_rate
print_section_divider()
# Step 6: Query batch size
print_batch_size_message()
batch_size = select_integer_value(print_msg=BATCH_SIZE_PRINT_MSG,
prompt_msg=BATCH_SIZE_PROMPT_MSG,
error_msg=BATCH_SIZE_ERROR_MSG)
config[BATCH_SIZE] = batch_size
print_section_divider()
# Step 7: Query number of epochs
print_number_epochs_message()
number_epochs = select_integer_value(print_msg=EPOCH_PRINT_MSG,
prompt_msg=EPOCH_PROMPT_MSG,
error_msg=EPOCH_ERROR_MSG)
config[NUM_EPOCHS] = number_epochs
print_section_divider()
return config
