cond_trainer.py

#This is only ever needed for non API models and somewhat outdated/deprecated. It is kept here for reference.
#Script that can be started in the background to train a prepared flow on prepared data, as specified in the cond_trainer folder.
#See Base_Processor_Workflow.ipynb for more information on how to prepare the data and model.
if __name__ == "__main__":
    import numpy as np
    import torch
    import flowcode
    import time
    import sys


    file_ext = sys.argv[1]
    train_data = torch.load("cond_trainer/data_cond_trainer.pth")
    device_specified = "cuda:" + sys.argv[2] if len(sys.argv) > 2 else None #GPU is either specified or not, if not model will be loaded on the device it was saved on.
    model = torch.load("cond_trainer/model_cond_trainer.pth", map_location=device_specified)
    train_parameters = np.load("cond_trainer/params_cond_trainer.npy")
    filename = str(np.load("cond_trainer/filename_cond_trainer.npy"))

    #Signal that loading is complete and the file can be overwritten safely
    np.save("cond_trainer/loading_complete.npy", np.array([1]))

    train_loss_saver = []

    start = time.perf_counter()

    flowcode.train_flow(model, train_data, train_parameters[:-4].tolist(), int(float(train_parameters[-4])), lr=float(train_parameters[-3]), batch_size=int(float(train_parameters[-2])), loss_saver=train_loss_saver, gamma=float(train_parameters[-1]), give_textfile_info=file_ext)

    end = time.perf_counter()

    torch.save(model.state_dict(), f"saves/{filename}.pth")
    np.save(f"saves/loss_{filename}.npy",np.array(train_loss_saver+[end-start]))