linear_model.ZINB_grad.train_ZINB

linear_model.ZINB_grad.train_ZINB(x, optimizer, model, epochs=150, val=False)[source]

Trains a ZINB-Grad model.

The function will train a ZINB-Grad model using an optimizer for a number of epochs,: and it will return both losses and negative log-likelihood, which were obtained during the training procedure.

Parameters

xtorch.Tensor: It is the data for training, a Tensor of shape (n_samples, n_features).
optimizer: An object of torch.optim.Optimizer: For more details, please refer to Pytorch documentation.
model: An object of the ZINB_Grad class: Please refer to the example.
epochsint (optional, default=150): Number of iteration for training.
valbool (optional, default=False): Whether it is validation or training process.

Returns

losseslist: A list consisting of the loss of each epoch.
neg_log_likslist: A list consisting of the negative Log-likelihood of each epoch.

Examples

>>> import ZINB_grad
>>> import data_prep
>>> import torch
>>> from torch.utils.data import DataLoader
>>> cortex = data_prep.CORTEX()
>>> y, labels = next(iter(DataLoader(cortex,
                             batch_size= cortex.n_cells,
                             shuffle=True)))
>>> model = ZINB_grad.ZINB_WaVE(Y = y, K = 10, device =device)
>>> optimizer = torch.optim.Adam(model.parameters(), lr = 0.08)
>>> losses, neg_log_liks = ZINB_grad.train_ZINB(y, optimizer, model, epochs = 300)