SCL/utils.py at master · dhh1995/SCL · 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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
#! /usr/bin/env python3
# -*- coding: utf-8 -*-
# File   : utils.py
# Author : Honghua Dong
# Email  : dhh19951@gmail.com
# Date   : 11/06/2019
#
# Distributed under terms of the MIT license.

import matplotlib.pyplot as plt

from scripts.utils import plot

__all__ = ['plot_curve', 'get_exp_name', 'get_image_title']


def plot_curve(all_meters, image_title, figure_file):
    train_acc = []
    test_acc = []
    for item in all_meters:
        train_meters, test_meters = item
        train_acc.append(train_meters['acc'].avg)
        test_acc.append(test_meters['acc'].avg)
    data_dict = dict(train_acc=train_acc, test_acc=test_acc)
    plot(data_dict, image_title, figure_file)


def get_task_abbv(task):
    names = []
    if len(task) > 1:
        names.append('joint')
    for t in task:
        name = t[:2]
        if t == 'in_distri':
            name += '4'
        if t == 'distribute_four':
            name += '4'
        if t == 'distribute_nine':
            name += '9'
        names.append(name)
    return '_'.join(names)


def get_name(prefix, arr, seperator='_'):
    name = ''
    for i, h in enumerate(arr):
        name += seperator
        if i == 0:
            name += '{}'.format(prefix)
        name += '{}'.format(h)
    return name


# get the exp name by (most frequently used) args
def get_exp_name(args):
    lr_str = '{}'.format(args.lr).replace('.', '')
    name = '{}_lr{}_nf{}_bs{}_ed{}_ne{}'.format(
        get_task_abbv(args.task), lr_str, args.num_features,
        args.batch_size, args.embedding_dim, args.num_experts)
    if args.v2s_lr is not None:
        name += '_vlr{}'.format(args.v2s_lr)
    if args.lr_anneal_start is not None:
        name += '_lr_s{}_i{}_r{}'.format(
            args.lr_anneal_start, args.lr_anneal_interval, args.lr_anneal_ratio)
    if args.observe_interval is not None:
        if args.obs_lr is not None:
            name += '_olr{}'.format(args.obs_lr)
    if args.normal_group_mlp:
        name += '_ngm'
    name += get_name('hd', args.hidden_dims)
    if args.feature_embedding_dim != 1:
        name += '_fed{}'.format(args.feature_embedding_dim)
    if args.use_visual_inputs:
        if args.prediction_beta != 1.0:
            name += '_pb{}'.format(args.prediction_beta)
        if args.symbolic_beta != 0.0:
            name += '_sb{}'.format(args.symbolic_beta)
        if args.image_size != [80, 80]:
            h, w = args.image_size
            name += '_is{}_{}'.format(h, w)
        if args.use_resnet:
            name += '_ur'
        if args.num_visual_experts > 1:
            name += '_nve{}'.format(args.num_visual_experts)
        if args.factor_groups > 1:
            name += '_fg{}'.format(args.factor_groups)
            if args.split_channel:
                name += '_sc'
        if args.transformed_spatial_dim is not None:
            name += '_tsd{}'.format(args.transformed_spatial_dim)
        if args.v2s_softmax:
            name += '_vss'
        name += get_name('chd', args.conv_hidden_dims)
        if args.conv_repeats is not None:
            name += get_name('cr', args.conv_repeats)
        if args.conv_residual_link:
            name += '_crl'
        name += get_name('vhd', args.visual_mlp_hidden_dims)
        name += get_name('thd', args.mlp_transform_hidden_dims)
    name += get_name('ehd', args.embedding_hidden_dims)
    name += get_name('lhd', args.lastmlp_hidden_dims)
    name += get_name('rg', args.reduction_groups)

    if args.sum_as_reduction > 0:
        name += '_sum{}'.format(args.sum_as_reduction)
    if args.one_hot:
        name += '_oh'
    if args.enable_residual_block:
        name += '_res'
    if args.weight_decay != 0.0:
        wd_str = '{}'.format(args.weight_decay).replace('.', '')
        name += '_wd{}'.format(wd_str)
    if args.use_layer_norm:
        name += '_ln'
    if args.adjust_size:
        name += '_adj'
    if args.not_use_softmax:
        name += '_ns'
    if args.exclude_angle_attr:
        name += '_ea'
    name += args.split
    if args.random_seed is not None:
        name += '_seed{}'.format(args.random_seed)
    if args.numpy_random_seed is not None:
        name += '_nseed{}'.format(args.numpy_random_seed)
    if args.torch_random_seed is not None:
        name += '_tseed{}'.format(args.torch_random_seed)
    if len(args.extra) > 0:
        name += '_' + args.extra
    return name


def get_image_title(args):
    title = args.image_title
    if title is None:
        title = '{},lr={},nf={},ed={},ne={},hd={},rg={}'.format(
            get_task_abbv(args.task), args.lr, args.num_features,
            args.embedding_dim, args.num_experts,
            args.hidden_dims, args.reduction_groups)
        if args.v2s_lr is not None:
            title += ',vlr{}'.format(args.v2s_lr)
        if args.weight_decay != 0.0:
            title += ',wd{}'.format(args.weight_decay)
        if args.lr_anneal_start is not None:
            title += ',lr s{}_r{}'.format(
                args.lr_anneal_start, args.lr_anneal_ratio)
            if args.lr_anneal_interval != 1:
                title += '_i{}'.format(args.lr_anneal_interval)
        if args.normal_group_mlp:
            title += ',ngm'
        if args.feature_embedding_dim != 1:
            title += ',fed{}'.format(args.feature_embedding_dim)
        if args.dataset_size is not None:
            title += ',ds{}'.format(args.dataset_size)
        if args.sum_as_reduction > 0:
            title += ',sum{}'.format(args.sum_as_reduction)
        if args.use_visual_inputs:
            if args.v2s_softmax:
                title += ',vss'
            if args.symbolic_beta != 0.0:
                title += ',sb{}'.format(args.symbolic_beta)
            if args.num_visual_experts > 1:
                title += ',nve{}'.format(args.num_visual_experts)
            if args.factor_groups > 1:
                title += ',fg{}'.format(args.factor_groups)
        if args.one_hot:
            title += ',oh'
        if args.enable_residual_block:
            title += ',res'
        if args.use_layer_norm:
            title += ',ln'
        if args.adjust_size:
            title += ',adj'
        if args.not_use_softmax:
            title += ',ns'
        if args.exclude_angle_attr:
            title += ',ea'
        # if args.use_visual_inputs:
        #     title += ',visual'

        split = args.split
        if len(split) > 0:
            split = ',' + split[1:]
            title += split.replace('_any', '').replace('_all', '')
    return title