Model Configuration¶
The Model Configuration file contains configuration for training the model, conducting data study, KMC Generation, Benchmarking and Transfer Learning
Model Training Parameters
- param model_parameters[‘model_type’]
The type of model to be used for training, currently defaults to 3D CNN
- type model_parameters[‘model_type’]
str (required)
- param model_parameters[‘output_type’]
(regression, classification) The output type of the model used to initialize the output layer, currently defaults to regression
- type model_parameters[‘output_type’]
str (required)
- param model_parameters[‘batch_size’]
The batch size while training, can be tuned based on the hardware specifications, currently defaults to 32
- type model_parameters[‘batch_size’]
int (required)
- param model_parameters[‘epocs’]
The number of epocs the model is to be trained for, currently defaults to 150
- type model_parameters[‘epocs’]
int (required)
- param model_parameters[‘split_ratio’]
Split Ratio for train and validation
- type model_parameters[‘split_ratio’]
float (required)
- param model_parameters[‘optimizer’]
The optimizer to be used for model training, refer https://keras.io/optimizers/ for more information, currently defaults to adam
- type model_parameters[‘optimizer’]
keras.optimizer (required)
- param model_parameters[‘loss_func’]
The loss function to be optimized while model training, refer https://keras.io/losses/ for more information, currently defaults to Mean Squared Error (MSE)
- type model_parameters[‘loss_func’]
keras.losses (required)
- param model_parameters[‘regularizer_coeff’]
The regularizing coefficient to be used for L2 norm regularization of the fully connected layer, refer https://keras.io/regularizers/ for more information currently defaults to 0.1
- type model_parameters[‘regularizer_coeff’]
float (required)
- param model_parameters[‘activate_tensorboard’]
Tensorboard activation flag https://www.tensorflow.org/tensorboard, currently set to 0, changes to 1 for activating tensorbiard, Warning: There can be some compatibility issues with different Tensorflow and Cuda Toolkit Versions
- type model_parameters[‘loss_func’]
int (required)
Data Study Parameters
- param data_study_params[‘batch_size’]
The batch size while conducting data study, can be tuned based on the hardware specifications, currently defaults to 32
- type data_study_params[‘batch_size’]
int (required)
- param data_study_params[‘epocs’]
The number of epocs the model is to be trained for, currently defaults to 50
- type data_study_params[‘epocs’]
int (required)
- param data_study_params[‘split_ratio’]
Split Ratio for train and validation during data study
- type data_study_params[‘split_ratio’]
float (required)
- param data_study_params[‘min_train_samples’]
Minimum train Samples for data study, currently defaults to 100
- type data_study_params[‘min_train_samples’]
int (required)
- param data_study_params[‘max_train_samples’]
Maximum train samples for data study, dataset size is the maximum value
- type data_study_params[‘max_train_samples’]
int (required)
- param data_study_params[‘train_increment’]
Increment in the train size with each iteration, currently defaults to 100
- type data_study_params[‘train_increment’]
int (required)
Key Measurment Characteristics Generation Parameters
- param kmc_params[‘tree_based_model’]
The model to be used while generating feature importance, refer: https://xgboost.readthedocs.io/en/latest/R-package/discoverYourData.html#measure-feature-importance for more details, currently defaults to xgb, random forests can also be used
- type kmc_params[‘tree_based_model’]
str (required)
- param kmc_params[‘tree_based_model’]
The importance criteria to be used, currently defaults to gini index
- type kmc_params[‘tree_based_model’]
str (required)
- param kmc_params[‘split_ratio’]
Split Ratio for train and validation during data study
- type kmc_params[‘split_ratio’]
float (required)
- param kmc_params[‘save_model’]
Flag to save the model, Currently defaults to 0, change to 1 if model needs to be saved
- type kmc_params[‘save_model’]
int (required)
- param kmc_params[‘plot_kmc’]
Flag to plot the KMC, Currently defaults to 1, change to 0 if no plotting is required
- type kmc_params[‘plot_kmc’]
int (required)
Benchmarking Parameters
- param bm_params[‘max_models’]
The maximum number of models to be used for benchmarking, currently defaults to 10
- type bm_params[‘max_models’]
int (required)
- param bm_params[‘runs’]
Number of benchmarking runs to be conducted
- type bm_params[‘runs’]
int (required)
Transfer Learning Parameters
- param transfer_learning[‘tl_type’]
The type of transfer learning to be used (full_fine_tune, variable_lr, feature_extractor) currently defaults to full_fine_tune
- type transfer_learning[‘tl_type’]
str (required)
- param transfer_learning[‘tl_base’]
The type of base model (3D CNN Architecture) to be used (pointdevnet, voxnet, 3d-UNet), currently defaults to PointdevNet
- type transfer_learning[‘tl_base’]
str (required)
- param transfer_learning[‘tl_app’]
The application of the transfer learning model (classification, regression), currently defaults to regression
- type transfer_learning[‘tl_app’]
str (required)
- param transfer_learning[‘conv_layer_m’]
The learning rate multiplier for convolution layers (only needed when tl_type is variable_lr), defaults to 0.1 (10% of the network Learning Rate)
- type transfer_learning[‘conv_layer_m’]
float (required)
- param transfer_learning[‘dense_layer_m’]
The learning rate multiplier for dense layers (only needed when tl_type is variable_lr), defaults to 1 (100% of the network Learning Rate)
- type transfer_learning[‘dense_layer_m’]
float (required)