Sumo-Net/debug_run.py at master · MrHuff/Sumo-Net · GitHub

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
from sumo_net.hyperopt_class import *
import numpy as np

import torch
import GPUtil
import warnings
warnings.simplefilter("ignore")
datasets = ['support',
            'metabric',
            'gbsg',
            'flchain',
            'kkbox',
            'weibull',
            'checkboard',
            'normal'
            ]
#Uppgrade dataloader rip, probably uses some retarded permutation which is really slow.
#Write serious job script, figure out post processing pipeline...
if __name__ == '__main__':
    #Evaluate other toy examples to draw further conclusions...
    # Time component might need to be normalized...
    # eval more frequently...
    hyper_param_space = {
        # torch.nn.functional.elu,torch.nn.functional.relu,
        'bounding_op': [torch.relu],  # torch.sigmoid, torch.relu, torch.exp,
        'transformation': [torch.nn.Tanh()],
        'depth_x': [4],
        'width_x': [16], #adapt for smaller time net
        'depth_t': [1],
        'width_t': [1], #ads
        'depth': [4],
        'width': [16],
        'bs': [50],
        'lr': [1e-2],
        'direct_dif': ['autograd'],
        'dropout': [0.1],
        'eps':[1e-4],
        'weight_decay':[0],
        'T_losses':[90],
        'alpha': [0.2],
        'sigma': [0.1],
        'num_dur': [20],

    }
    for i in [0]:
        devices = GPUtil.getAvailable(order='memory', limit=8)
        print(devices)
        print(torch.cuda.device_count())
        device = devices[0]
        job_params = {
            'd_out': 1,
            'dataset_string': datasets[i],
            'seed': 1,#,np.random.randint(0,9999),
            'total_epochs': 50,
            'device': device,
            'patience': 50,
            'hyperits': 5,
            'selection_criteria':'train',
            'grid_size':100,
            'test_grid_size':100,
            'validation_interval':2,
            # 'net_type':'survival_net_basic',
            # 'net_type':'weibull_net',
            # 'net_type':'lognormal_net',
            'net_type':'survival_net_basic',
            # 'net_type':'cox_time_benchmark',
            # 'net_type':'deephit_benchmark',
            # 'net_type':'cox_linear_benchmark',
            # 'net_type':'deepsurv_benchmark',
            'objective': 'S_mean',
            'fold_idx':1 ,
            'savedir':'test',
            'use_sotle':False,
        }
        training_obj = hyperopt_training(job_param=job_params,hyper_param_space=hyper_param_space)
        # training_obj.debug=True
        training_obj.run()
        training_obj.post_process()