retrain model

[gerrardmai]

Could I retain the existed model with new data?
any function could provided?

[Shao, Xin]

Yes, the function is train.py

[gerrardmai]

Yes, the function is train.py

but i cannot find any parser.add_argument to load the existed model?

#######
parser = argparse.ArgumentParser()
parser.add_argument("--random_seed", type=int, default=10086)
parser.add_argument("--dropout", type=float, default=0.1,
help="dropout probability")
parser.add_argument("--gpu", type=int, default=2,
help="GPU id, -1 for cpu")
parser.add_argument("--filetype", default='csv', type=str, choices=['csv', 'gz'],
help='data file type, csv or gz')
parser.add_argument("--lr", type=float, default=1e-3,
help="learning rate")
parser.add_argument("--weight_decay", type=float, default=5e-4,
help="Weight for L2 loss")
parser.add_argument("--n_epochs", type=int, default=10,
help="number of training epochs")
parser.add_argument("--dense_dim", type=int, default=400,
help="number of hidden gcn units")
parser.add_argument("--hidden_dim", type=int, default=200,
help="number of hidden gcn units")
parser.add_argument("--n_layers", type=int, default=1,
help="number of hidden gcn layers")
parser.add_argument("--threshold", type=float, default=0,
help="the threshold to connect edges between cells and genes")
parser.add_argument("--num_neighbors", type=int, default=0,
help="number of neighbors to sample in message passing process. 0 means all neighbors")
parser.add_argument("--exclude_rate", type=float, default=0,
help="exclude some cells less than this rate.")
parser.add_argument("--species", default='mouse', type=str)
parser.add_argument("--tissue", required=True, type=str)
parser.add_argument("--batch_size", type=int, default=500)
parser.add_argument("--unsure_rate", type=float, default=2.,
help="the threshold to predict unsure cell")
parser.add_argument("--test_rate", type=float, default=0.2)

[Shao, Xin]

Hello, the train.py is to train a new model and it will replace the pre-trained model if the new model has the same tissue name as the one in pre-trained models. You want a model trained with our training data and your new data?

[gerrardmai]

Hello, the train.py is to train a new model and it will replace the pre-trained model if the new model has the same tissue name as the one in pre-trained models. You want a model trained with our training data and your new data?

Hello,
my case is I have trained my own model already .
For example the model is m1 which trained by dataset1. now I want to train the m1 (which is existed ) with other new dataset2.
I know I can merge dataset1 and dataset2 into one csv, then train the model with merged dataset. But it may take times.

[Shao, Xin]

Sorry that the current version of scDeepSort doesn't have this function. You can merge dataset1 and dataset2 into one csv, then train the model though it may take time.