olfaction_evolution/peter_cluster.py at master · gyyang/olfaction_evolution · 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
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
#changed
from collections import OrderedDict
from collections.__init__ import OrderedDict

import argparse
import configs
import standard.experiment as se
from standard.hyper_parameter_train import local_train, cluster_train

parser = argparse.ArgumentParser()
parser.add_argument('-p','--pn', nargs='+', help='N_PN', default=[50])
args = parser.parse_args()

def temp(n_pn=50):
    config = configs.FullConfig()
    config.N_PN = n_pn
    config.data_dir = './datasets/proto/orn' + str(n_pn)

    config.max_epoch = 100
    config.direct_glo = True

    config.lr = 3e-3

    config.kc_dropout = True
    config.kc_dropout_rate = 0.5

    config.train_pn2kc = True
    config.sparse_pn2kc = False
    config.coding_level = None

    config.save_log_only = True

    config.initial_pn2kc = 5 / n_pn
    config.kc_prune_weak_weights = False

    # Ranges of hyperparameters to loop over
    hp_ranges = OrderedDict()
    hp_ranges['lr'] = [3e-3, 1e-3, 3e-4, 1e-4, 3e-5, 1e-5]
    return config, hp_ranges

def temp_(n_pn=50):
    config = configs.FullConfig()
    config.N_PN = n_pn
    config.data_dir = './datasets/proto/orn' + str(n_pn)

    config.max_epoch = 50
    config.direct_glo = True

    config.lr = 3e-3

    config.kc_dropout = True
    config.kc_dropout_rate = 0.5

    config.train_pn2kc = True
    config.sparse_pn2kc = False
    config.coding_level = None

    config.save_log_only = True

    config.initial_pn2kc = 6 / n_pn
    config.kc_prune_threshold = 5 / n_pn
    config.kc_prune_weak_weights = True

    # Ranges of hyperparameters to loop over
    hp_ranges = OrderedDict()
    hp_ranges['lr'] = [3e-3, 1e-3, 3e-4, 1e-4, 3e-5, 1e-5]
    hp_ranges['N_KC'] = [2500, 5000, 10000]
    hp_ranges['kc_prune_threshold'] = [1/n_pn, 2/n_pn, 5/n_pn]
    hp_ranges['kc_dropout_rate'] = [0, .3, .6]
    # hp_ranges['initial_pn2kc'] = [2/n_pn, 5/n_pn, 10/n_pn]
    return config, hp_ranges

def temp_glomeruli(n_pn=50):
    config = configs.FullConfig()
    config.N_PN = n_pn
    config.data_dir = './datasets/proto/orn' + str(n_pn)

    config.max_epoch = 100
    config.pn_norm_pre = 'batch_norm'

    config.train_pn2kc = False
    config.sparse_pn2kc = True

    config.save_log_only = False

    config.initializer_orn2pn = 'constant'
    config.initial_orn2pn = .1
    config.pn_prune_threshold = .05

    # Ranges of hyperparameters to loop over
    hp_ranges = OrderedDict()
    hp_ranges['pn_prune_weak_weights'] = [True, False]
    return config, hp_ranges

def rnn(n_pn=50):
    config = configs.FullConfig()
    config.data_dir = './datasets/proto/orn' + str(n_pn)
    config.max_epoch = 8
    config.model = 'rnn'
    config.lr= 1e-3

    config.NEURONS = 2500
    config.WEIGHT_LOSS = False
    config.WEIGHT_ALPHA = 0
    config.BATCH_NORM = False
    config.DIAGONAL_INIT = True

    config.dropout = True
    config.dropout_rate = .5
    config.save_every_epoch = False

    config.initial_rnn_weight = .1
    config.prune_threshold = .01
    config.prune_weak_weights = True

    hp_ranges = OrderedDict()
    # hp_ranges['TIME_STEPS'] = [1, 2, 3]
    # hp_ranges['replicate_orn_with_tiling'] = [False, True, True]
    # hp_ranges['N_ORN_DUPLICATION'] = [1, 10, 10]

    hp_ranges['TIME_STEPS'] = [2]
    hp_ranges['replicate_orn_with_tiling'] = [True]
    hp_ranges['N_ORN_DUPLICATION'] = [5]
    return config, hp_ranges

train = cluster_train
cluster_path = '/share/ctn/users/yw2500/olfaction_evolution'
n_pns = [int(x) for x in args.pn]
print(n_pns)
for n_pn in n_pns:
    path = './files/cluster_simple_no_prune' + str(n_pn)
    cluster_train(temp(n_pn), save_path=path, sequential=False, path= cluster_path)
#
# n_pns = [50]
# for n_pn in n_pns:
#    path = './files/test' + str(n_pn)
#    try:
#        import shutil
#        shutil.rmtree(path)
#    except:
#        pass
#    local_train(rnn(n_pn), sequential=True, save_path=path)