HoloNet.predicting.mgc_repeat_training

HoloNet.predicting.mgc_repeat_training(X, adj, target, repeat_num=50, train_set_ratio=0.85, val_set_ratio=0.15, hidden_num=None, max_epoch=300, lr=0.1, weight_decay=0.0005, step_size=10, gamma=0.9, display_loss=False, only_cell_type=False, hide_repeat_tqdm=False, device='cpu')

Using cell-type tensor and normalized adjancency matrix as the inputs, repeated training GNN to generate the target gene expression.

Parameters

X Tensor

A tensor (cell_num * cell_type_num) with cell-type information. derived from ‘get_continuous_cell_type_tensor’ or ‘get_one_hot_cell_type_tensor’ function.

adj Tensor

A normalized adjancency matrix derived from ‘adj_normalize’ function.

target Tensor

The scaled expression tensor of one target gene (cell_num * 1), derived from ‘get_one_case_expr’ function.

repeat_num int (default: 50)

The number of repeated training, defaultly as 50.

train_set_ratio float (default: 0.85)

A value from 0-1. The ratio of cells using as the training set.

val_set_ratio float (default: 0.15)

A value from 0-1. The ratio of cells using as the validation set.

hidden_num int | NoneOptional[int] (default: None)

The dim of ‘MultiGraphConvolution_Layer’ output. Always use 1 or same as feature_num.

max_epoch int (default: 300)

The maximum epoch of training/

lr float (default: 0.1)

The learning rate.

weight_decay float (default: 0.0005)

The weight decay (L2 penalty)

step_size int (default: 10)

Period of learning rate decay.

gamma float (default: 0.9)

Multiplicative factor of learning rate decay.

display_loss bool (default: False)

If true, display the loss during training.

only_cell_type bool (default: False)

If true, the model only use the Feature matrix training target, serving as a baseline model.

hide_repeat_tqdm bool (default: False)

If true, hide the tqdm for repeated training.

use_gpu

If true, model will be trained in GPU when GPU is available.

device str (default: 'cpu')

Give a device to use

Return type

List[MGC_Model]

Returns

: A list of trained MGC model for generating the expression of one target gene.