If the program could not connect to Visdom server, this function will start a new server at the default port.

Unpack input data from the dataloader and perform necessary pre-processing steps. The implementation here is just a basic setting of input and

Create a dataset given the configuration (loaded from the json file). This function wraps the class CustomDatasetDataLoader. This is the

Make models eval mode during test time.

Calculate losses; called in every training iteration.

Create a model given the configuration. This is the main interface between this package and train.py/validate.py

Run forward pass; called by both functions <optimize_parameters> and <test>.

Loads config file if a string was passed and returns the input if a dictionary was passed.

Load and print networks; create schedulers.

Forward function used in test time. This function wraps <forward> function in no_grad() so we don't save intermediate steps for backprop

Transfers pytorch tensors or lists of tensors to GPU. This function is recursive to be able to deal with lists of lists.

Exports all the networks of the model using JIT tracing. Requires that the input is set.

Import the module "data/[dataset_name]_dataset.py". In the file, the class called DatasetNameDataset() will be instantiated. It has to be a s

Import the module "models/[model_name]_model.py". In the file, the class called DatasetNameModel() will be instantiated. It has to be

Return traning losses / errors. train.py will print out these errors on console

Get a custom dataloader (e.g. for exporting the model). This dataloader may use different configurations than the default trai

Returns the final training result for hyperparameter tuning (e.g. best validation loss).

Return a learning rate scheduler.

Load all the networks from the disk.

Load all the optimizers from the disk.

Calculate losses, gradients, and update network weights; called in every training iteration

Display the current losses on visdom display: dictionary of error labels and values. Input params: epoch: Current epoch.

Display the current validation metrics on visdom display: dictionary of error labels and values. Input params: epoch: Current epo

Print current losses on console. Input params: epoch: Current epoch. max_epochs: Maximum number of epochs.

Print the total number of parameters in the network and network architecture.

Reset the visualization.

Save all the networks to the disk.

Save all the optimizers to the disk for restarting training.

Make models train mode during test time.

Update learning rates for all the networks; called at the end of every epoch

Return a data point (usually data and labels in a supervised setting).

Initialize the Visualizer class. Input params: configuration -- stores all the configurations

Initialize the class; save the configuration in the class.

Initialize the BaseModel class. Parameters: configuration: Configuration dictionary. When creating your custom class, yo

Initialize the model.

Return a batch of data.

Return the total number of images in the dataset.

Return the number of data in the dataset.

Run forward pass.

Return visualization images. train.py will display these images.

Calculate gradients and update network weights.

Display the ROC curve. Input params: fpr: False positive rate (1 - specificity). tpr: True positive rate (sensitivity

Callback to be called after every epoch.

Callback to be called before every epoch.

Set requies_grad for all the networks to avoid unnecessary computations.

Display validation images. The images have to be in the form of a tensor with [(image, label, prediction), (image, label, prediction), ...] in

Stack all iterables of torch tensors in a list (i.e. [[(tensor), (tensor)], [(tensor), (tensor)]])