import argparse import datetime import logging import numpy as np impo

  • Anonymous
  • Python
  • 01 Mar 2019 
  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
import argparse

import datetime

import logging

import numpy as np

import operator

import os

import pandas as pd

import re

import shutil

import sys



import torchvision

from tensorboardX import SummaryWriter

from time import gmtime, strftime



import torch

from tqdm import tqdm

from torch.utils.data import DataLoader

from torchvision import transforms



from src.data_loader import MakeDataSynt

from src.loss import MSE_synthetic_loss

from src.models import DLD





def parse_args():

    parser = argparse.ArgumentParser()



    parser.add_argument('--model', type=str, required=True, help='What model to use, one of ["DLD", ]', default='DLD')

    parser.add_argument('--n_epochs', type=int, help='Num of epochs for training', default=10)

    parser.add_argument('--datadir', type=str, help='Path to training dataset')

    parser.add_argument('--valdatadir', type=str, help='Path to validation dataset')

    parser.add_argument('--batch_size', type=int, default=8)

    parser.add_argument('--name', type=str, help='Name of the experiment')



    args = parser.parse_args()

    return args





def get_dataloaders(args):

    train_transform = transforms.Compose([

        transforms.ToTensor(),

        transforms.ColorJitter()

    ])



    dset_synt = MakeDataSynt(args.datadir, args.datadir, train_transform)

    dset_val_synt = MakeDataSynt(args.valdatadir, args.valdatadir, train_transform)



    train_loader = DataLoader(dset_synt,

                              batch_size=1,

                              shuffle=True,

                              num_workers=0,  # 1 for CUDA

                              pin_memory=False)



    val_loader = DataLoader(dset_val_synt,

                            batch_size=1,

                            shuffle=True,

                            num_workers=0,  # 1 for CUDA

                            pin_memory=False)



    return train_loader, val_loader





def validate(tb, val_loader, model, device, global_step):

    val_loss_epoch = []



    for X_batch, y_batch_h, y_batch_nh in tqdm(val_loader):

        X_batch = torch.FloatTensor(X_batch).to(device)

        y_batch_nh = y_batch_nh.type(torch.FloatTensor).to(device)

        y_batch_h = y_batch_h.type(torch.FloatTensor).to(device)



        loss = MSE_synthetic_loss(X_batch, y_batch_nh, y_batch_h, model, device)

        val_loss_epoch.append(loss.cpu().data.numpy())

        del loss



    tb.add_scalar('val_loss', np.mean(val_loss_epoch), global_step=global_step)



    out = model(X_batch)[0]



    



    out_grid = torchvision.utils.make_grid(out.cpu())

    input_grid = torchvision.utils.make_grid(X_batch.cpu())



    tb.add_image(tag='val_out', img_tensor=out_grid, global_step=global_step)

    tb.add_image(tag='val_input', img_tensor=input_grid, global_step=global_step)







def main(args):

    tb_dir = './tb_logs/' + args.name + ' ' + strftime("%Y-%m-%d %H:%M:%S", gmtime())

    tb = SummaryWriter(tb_dir)



    device = torch.device("cuda:0")



    train_loader, val_loader = get_dataloaders(args)



    model = DLD().to(device)

    # model.load_state_dict(torch.load('./model.pth'))

    model.eval()



    tb.add_text(tag='model', text_string=repr(model))



    opt = torch.optim.Adam(model.parameters(), lr=0.001)



    global_step = 0



    for epoch in range(args.n_epochs):



        validate(tb, val_loader, model, device, global_step=global_step)



        model.train(True)

        for X_batch, y_batch_h, y_batch_nh in tqdm(train_loader):

            X_batch = torch.FloatTensor(X_batch).to(device)

            y_batch_nh = y_batch_nh.type(torch.FloatTensor).to(device)

            y_batch_h = y_batch_h.type(torch.FloatTensor).to(device)



            loss = MSE_synthetic_loss(X_batch, y_batch_nh, y_batch_h, model, device)

            loss.backward()

            opt.step()

            opt.zero_grad()



            tb.add_scalar('train_loss', loss.cpu().data.numpy(), global_step=global_step)



            global_step += 1



        out = model(X_batch)[0]



        out_grid = torchvision.utils.make_grid(out.cpu())

        input_grid = torchvision.utils.make_grid(X_batch.cpu())



        tb.add_image(tag='train_out', img_tensor=out_grid, global_step=global_step)

        tb.add_image(tag='train_input', img_tensor=input_grid, global_step=global_step)



        model.train(False)









if __name__ == '__main__':

    args = parse_args()

    main(args)
← newer dump
older dump →