Exemple #1
0
#!/usr/bin/env python
import os, sys
import torch
import setlog

conf_file = os.environ['DEV'] + 'dl_management/.log/logging.yaml'
save_file = os.path.abspath(sys.argv[0])[:-len(sys.argv[0])] + 'log/'
setlog.reconfigure(conf_file, save_file)

import system.DescriptorLearning as System

if __name__ == '__main__':
    machine = System.Default(
        root=os.path.abspath(sys.argv[0])[:-len(sys.argv[0])])
    action = input('''
    Exec:
    [t]\ttrain
    [e]\ttest
    [p]\tprint (console)
    [P]\tprint (full)
    [s]\tserialize net
    [c]\tcreat clusters
    [ ]\ttrain+test
    ''')
    if action == 't':
        machine.train()
    elif action == 'e':
        machine.test()
        machine.plot(print_loss=False, print_val=False)
    elif action == 'p':
        machine.plot(print_loss=False, print_val=False)
Exemple #2
0
#!/usr/bin/env python
import os, sys
import setlog

conf_file = os.environ['DEV'] + 'dl_management/.log/logging.yaml'
save_file = os.path.abspath(sys.argv[0])[:-len(sys.argv[0])] + 'log/'
setlog.reconfigure(conf_file, save_file)

import system.DescriptorLearning as System

if __name__ == '__main__':
    machine = System.Default(root=os.path.abspath(
        sys.argv[0])[:-len(sys.argv[0])],
                             dataset_file='../../../../datasets/default.yaml',
                             trainer_file='../trainer.yaml')
    action = input(
        'Exec:\n[t]\ttrain\n[e]\ttest\n[p]\tprint (console)\n[P]\tprint (full)\n[ ]\ttrain+test\n'
    )
    if action == 't':
        machine.train()
    elif action == 'e':
        machine.test()
        machine.plot(print_loss=False, print_val=False)
    elif action == 'p':
        machine.plot(print_loss=False, print_val=False)
    elif action == 'P':
        machine.plot()
    elif action == '':
        machine.train()
        machine.test()
        machine.plot(print_loss=False, print_val=False)