mechanoChemML.workflows.mr_learning.mrnn_models

Module Contents

Classes

callback_PrintDot	Display training progress by printing a single dot for each completed epoch
user_DNN_kregl1l2_gauss_grad	DNN with regularization layers and Gaussian noise layers.
CNN_user_supervise	similar as CNN supervise, but with the check_layer() function

Functions

build_learningrate(config)	Build different learning rates based on the input
build_optimizer(config)	Build different optimizers based on the input
check_point_callback(config)	Return the check point call back function
tensor_board_callback(config)	Return the tensor board call back function
build_callbacks(config)	Build different call back functions based on the input
my_mse_loss()	Use defined mse loss function
my_mse_loss_with_grad(BetaP=1000.0)	Use defined mse loss function with penalty term for physics-based constraint
build_loss(config, loss_model=None)	Build loss function based on the input
shift_labels(config, dataset, dataset_index, dataset_frame, data_file)	Shift label based on the trained-NN predictions
load_one_model(old_config_file)	Based on the config file name to load pre-saved model info
load_trained_model(trained_model_lists)	Load trained models
user_DNN_kregl1l2_gauss_grad_setup(config, train_dataset, train_labels, NodesList, Activation, train_stats)
DNN_kregl1l2_gauss(config, train_dataset, train_labels, NodesList, Activation)	Customized DNN with different regularizations
add_input_layer(config, model, train_dataset, Name, Node, Act, padding=’valid’)	Add different layers to the model
add_one_layer(config, model, Name, Node, Act, padding=’valid’, tf_name=’’)	Add different layers to the model
add_output_layer(config, model, train_labels)	Add an output layer to the model
CNN_supervise(config, train_dataset, train_labels, NodesList, Activation, LayerName, Padding)	Generate a supervised CNN model based on the inputs
CNN_user_supervise_setup(config, train_dataset, train_labels, NodesList, Activation, LayerName, Padding)	Generate a supervised CNN model based on the inputs
build_model(config, train_dataset, train_labels, set_non_trainable=False, train_stats=None)	Build the NN model based on the input

mechanoChemML.workflows.mr_learning.mrnn_models.build_learningrate(config)

Build different learning rates based on the input

mechanoChemML.workflows.mr_learning.mrnn_models.build_optimizer(config)

Build different optimizers based on the input

class mechanoChemML.workflows.mr_learning.mrnn_models.callback_PrintDot

Bases: tensorflow.keras.callbacks.Callback

Display training progress by printing a single dot for each completed epoch

on_epoch_end(self, epoch, logs)

mechanoChemML.workflows.mr_learning.mrnn_models.check_point_callback(config)

Return the check point call back function

mechanoChemML.workflows.mr_learning.mrnn_models.tensor_board_callback(config)

Return the tensor board call back function

mechanoChemML.workflows.mr_learning.mrnn_models.build_callbacks(config)

Build different call back functions based on the input

mechanoChemML.workflows.mr_learning.mrnn_models.my_mse_loss()

Use defined mse loss function

mechanoChemML.workflows.mr_learning.mrnn_models.my_mse_loss_with_grad(BetaP=1000.0)

Use defined mse loss function with penalty term for physics-based constraint

mechanoChemML.workflows.mr_learning.mrnn_models.build_loss(config, loss_model=None)

Build loss function based on the input

mechanoChemML.workflows.mr_learning.mrnn_models.shift_labels(config, dataset, dataset_index, dataset_frame, data_file)

Shift label based on the trained-NN predictions

mechanoChemML.workflows.mr_learning.mrnn_models.load_one_model(old_config_file)

Based on the config file name to load pre-saved model info

mechanoChemML.workflows.mr_learning.mrnn_models.load_trained_model(trained_model_lists)

Load trained models

class mechanoChemML.workflows.mr_learning.mrnn_models.user_DNN_kregl1l2_gauss_grad(config, NodesList, Activation, train_dataset, train_labels, label_scale, train_stats)

Bases: tensorflow.keras.Model

DNN with regularization layers and Gaussian noise layers.

call(self, inputs)

class mechanoChemML.workflows.mr_learning.mrnn_models.CNN_user_supervise(input_shape, num_output, LayerName, NodesList, Activation, Padding, OutAct)

Bases: tensorflow.keras.Model

similar as CNN supervise, but with the check_layer() function

call(self, inputs, training=False)

check_layer(self, inputs, layer_number=100)

mechanoChemML.workflows.mr_learning.mrnn_models.user_DNN_kregl1l2_gauss_grad_setup(config, train_dataset, train_labels, NodesList, Activation, train_stats)

mechanoChemML.workflows.mr_learning.mrnn_models.DNN_kregl1l2_gauss(config, train_dataset, train_labels, NodesList, Activation)

Customized DNN with different regularizations

mechanoChemML.workflows.mr_learning.mrnn_models.add_input_layer(config, model, train_dataset, Name, Node, Act, padding='valid')

Add different layers to the model

mechanoChemML.workflows.mr_learning.mrnn_models.add_one_layer(config, model, Name, Node, Act, padding='valid', tf_name='')

Add different layers to the model

mechanoChemML.workflows.mr_learning.mrnn_models.add_output_layer(config, model, train_labels)

Add an output layer to the model

mechanoChemML.workflows.mr_learning.mrnn_models.CNN_supervise(config, train_dataset, train_labels, NodesList, Activation, LayerName, Padding)

Generate a supervised CNN model based on the inputs

mechanoChemML.workflows.mr_learning.mrnn_models.CNN_user_supervise_setup(config, train_dataset, train_labels, NodesList, Activation, LayerName, Padding)

Generate a supervised CNN model based on the inputs

mechanoChemML.workflows.mr_learning.mrnn_models.build_model(config, train_dataset, train_labels, set_non_trainable=False, train_stats=None)

Build the NN model based on the input