fastvpinns.model.model_hard module
The file model_hard.py contains the DenseModel class which is a custom model for the Neural Network for solving Variational PINNs. This model is used for enforcing hard boundary constraints on the solution.
Author: Thivin Anandh D, Divij Ghose, Sashikumaar Ganesan
Date: 22/Jan/2024
Changelog: 22/Jan/2024 - file created
Known issues: None
- class fastvpinns.model.model_hard.DenseModel_Hard(*args, **kwargs)[source]
Bases:
ModelThe DenseModel_Hard class is a custom model class that hosts the neural network model.
The class inherits from the tf.keras.Model class and is used to define the neural network model architecture and the training loop for FastVPINNs.
- Parameters:
layer_dims (list) – List of integers representing the number of neurons in each layer
learning_rate_dict (dict) – Dictionary containing the learning rate parameters
params_dict (dict) – Dictionary containing the parameters for the model
loss_function (function) – Loss function for the model
input_tensors_list (list) – List of input tensors for the model
orig_factor_matrices (list) – List of original factor matrices
force_function_list (list) – List of force functions
tensor_dtype (tf.DType) – Tensor data type
use_attention (bool) – Flag to use attention layer
activation (str) – Activation function for the model
hessian (bool) – Flag to compute hessian
- train_step(beta=10, bilinear_params_dict=None)[source]
This method is used to define the training step of the model.
- call(inputs)[source]
This method is used to define the forward pass of the model.
- Parameters:
inputs (tf.Tensor) – Input tensor
- Returns:
Output tensor from the model
- Return type:
tf.Tensor