-
Notifications
You must be signed in to change notification settings - Fork 2
/
test.py
303 lines (244 loc) · 11.9 KB
/
test.py
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
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
#Common imports
import sys
import os
import argparse
import random
import copy
import seaborn
import pandas as pd
import pickle
import matplotlib
import matplotlib.pyplot as plt
import torch
from torch import nn, optim
import torch.utils.data as data_utils
import torchvision
from torchvision.utils import save_image
from torch.autograd import Variable
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
from scipy import stats
from sklearn.decomposition import FastICA
#Algorithms
from algorithms.base_auto_encoder import AE
from algorithms.additive_auto_encoder import AE_Additive
#DataLoaders
from data.balls_dataset_loader import BallsDataLoader
from data.balls_dataset_loader import sample_base_data_loaders, sample_latent_traversal_data
#Metrics
from utils.metrics import *
# Input Parsing
parser = argparse.ArgumentParser()
parser.add_argument('--method_type', type=str, default='ae_additive',
help= 'ae, ae_poly')
parser.add_argument('--latent_case', type=str, default='balls_supp_l_shape',
help='laplace; uniform')
parser.add_argument('--eval_latent_case', type=str, default='balls_supp_l_shape',
help='laplace; uniform')
parser.add_argument('--plot_case', type=str, default='extrapolate',
help= 'extrapolation, block_rendering, latent_traversal')
parser.add_argument('--data_dim', type=int, default= 200,
help='')
parser.add_argument('--latent_dim', type=int, default= 2,
help='')
parser.add_argument('--total_blocks', type=int, default= 2,
help='')
parser.add_argument('--train_size', type=int, default= 50000,
help='')
parser.add_argument('--batch_size', type=int, default= 64,
help='')
parser.add_argument('--lr', type=float, default= 5e-4,
help='')
parser.add_argument('--weight_decay', type=float, default= 5e-4,
help='')
parser.add_argument('--num_seeds', type=int, default=10,
help='')
parser.add_argument('--target_seed', type=int, default=-1,
help='')
parser.add_argument('--input_normalization', type=str, default='none',
help= '')
parser.add_argument('--wandb_log', type=int, default=0,
help='')
parser.add_argument('--cuda_device', type=int, default=0,
help='Select the cuda device by id among the avaliable devices' )
args = parser.parse_args()
method_type= args.method_type
latent_case= args.latent_case
plot_case= args.plot_case
total_blocks= args.total_blocks
eval_latent_case= args.eval_latent_case
data_dim= args.data_dim
latent_dim= args.latent_dim
train_size= args.train_size
batch_size= args.batch_size
lr= args.lr
weight_decay= args.weight_decay
num_seeds= args.num_seeds
target_seed= args.target_seed
plot_case= args.plot_case
input_normalization= args.input_normalization
wandb_log= args.wandb_log
cuda_device= args.cuda_device
#GPU
if cuda_device == -1:
device= "cpu"
else:
device= torch.device("cuda:" + str(cuda_device))
if device:
kwargs = {'num_workers': 0, 'pin_memory': False}
else:
kwargs= {}
res={}
print('Details')
print(method_type, eval_latent_case, train_size)
for seed in range(num_seeds):
if target_seed != -1 and seed!=target_seed:
continue
#Seed values
random.seed(seed*10)
np.random.seed(seed*10)
torch.manual_seed(seed*10)
# Load Dataset
eval_batch_size= 10000
train_dataset, val_dataset, test_dataset= sample_base_data_loaders(
latent_case= eval_latent_case,
num_balls= total_blocks,
train_size= train_size,
batch_size= eval_batch_size,
input_normalization= input_normalization,
kwargs=kwargs
)
#Load Algorithm
if method_type == 'ae_base':
method= AE(args, train_dataset, val_dataset, test_dataset, seed=seed, device= device)
elif method_type == 'ae_additive':
method= AE_Additive(args, train_dataset, val_dataset, test_dataset, seed=seed, device= device)
else:
print('Error: Incorrect method type')
sys.exit(-1)
# Evaluate the base model
method.load_model()
#Obtain Predictions and Reconstruction Loss
save_dir= 'results/' + input_normalization + '/' + args.latent_case + '_eval_latent_' + args.eval_latent_case + \
'/' + args.method_type + '/' + str(args.train_size) + '/'
if not os.path.exists(save_dir):
os.makedirs(save_dir)
if method_type == 'ae_additive':
logs= method.eval_identification(seed= seed, save_dir= save_dir, plot= True)
else:
logs= method.eval_identification(seed= seed, save_dir= save_dir, plot= False)
#Store metrics over different random seeds
for key in ['recon_rmse', 'mcc_pearson', 'mcc_spearman', 'mcc_block']:
if key not in res.keys():
res[key]= []
res[key].append(logs[key])
#Plot the latent support and the images for the latent cartesian-product extrapolation for the ScalarLatentsDataset
if plot_case == 'extrapolate':
fontsize= 30
fontsize_lgd= fontsize/1.5
pred_latent= logs['pred_z']
true_latent= logs['true_z']
width= 64
height= 64
padding= 10
grid_size= 7
final_image= Image.new('RGB', ( 3*padding + (width+padding)*grid_size, 3*padding + (height+padding)*grid_size))
init_val= np.min(pred_latent[:, 0])
final_val= np.max(pred_latent[:, 0])
grid_range_x= np.linspace(init_val, final_val, num=grid_size)
init_val= np.min(pred_latent[:, 1])
final_val= np.max(pred_latent[:, 1])
grid_range_y= np.linspace(init_val, final_val, num=grid_size)
for idx_x in range(grid_size):
supp_val= grid_range_x[idx_x]
latent_traverse= []
for val in grid_range_y:
latent_traverse.append([supp_val, val])
latent_traverse= torch.tensor( np.array(latent_traverse) ).float()
x_pred= method.decoder( latent_traverse.to(device) ).to('cpu').detach()
for idx_y in range(x_pred.shape[0]):
data= x_pred[idx_y]
data= plot_transform(data, input_normalization= input_normalization)
data= torchvision.transforms.functional.to_pil_image(data)
final_image.paste(data, ( 2*padding+ (width+padding) * idx_x, 2*padding + (height+padding)*(grid_size - idx_y -1)))
final_image.save(save_dir + 'traversed_image_seed_' + str(seed) + '.jpg')
color_latent= []
for idx in range(pred_latent.shape[0]):
color_latent.append( true_latent[idx, 0] )
plt.scatter(pred_latent[:, 0], pred_latent[:, 1], c=color_latent)
plt.xlabel('Predicted Latent 1', fontsize= fontsize)
plt.ylabel('Predicted Latent 2', fontsize= fontsize)
ood_x= []
ood_y= []
for idx_x in range(grid_size):
for idx_y in range(grid_size):
ood_x.append(grid_range_x[idx_x])
ood_y.append(grid_range_y[idx_y])
ood_x= np.array(ood_x)
ood_y= np.array(ood_y)
plt.scatter(ood_x, ood_y, c='red')
plt.xticks(fontsize=18)
plt.yticks(fontsize=18)
plt.tight_layout()
plt.savefig(save_dir + 'latent_support_seed_' + str(seed) + '.jpg')
plt.clf()
print('Done')
#Plot the individual latent responses for the BlockLatentsDataset for sparse changes in the true latent
elif plot_case == 'latent_traversal':
x_ticks= 0.1 + 0.1 * np.array( range(9) )
fontsize=48
fontsize_lgd= fontsize/1.25
marker_list = ['o', '^', 's', '*']
fig, ax = plt.subplots(2, 2, figsize=(23, 16))
movement_axis= ['x', 'y']
for idx, axis in enumerate(movement_axis):
if axis == 'x':
latent_traversal_case= 'balls_latent_traversal_x_axis'
elif axis == 'y':
latent_traversal_case= 'balls_latent_traversal_y_axis'
train_loader = sample_latent_traversal_data(
latent_case= latent_traversal_case,
num_balls= total_blocks,
input_normalization= input_normalization,
kwargs=kwargs
)
with torch.no_grad():
for batch_idx, (x, z) in enumerate(train_loader):
pred_latent= method.encoder(x.to(device)).to('cpu').detach().view(9, 9, latent_dim).numpy()
curr_latent= pred_latent[:, 4, :]
ax[idx, 0].grid(True)
ax[idx, 0].set_ylim(-2.50, 2.50)
# if latent_traversal_case == 'balls_latent_traversal_y_axis':
# ax[idx, 0].set_ylim(-3.50, 2.50)
ax[idx, 0].tick_params(labelsize=fontsize)
ax[idx, 0].set_ylabel('Predicted Latents', fontsize=fontsize)
ax[idx, 0].set_xlabel('Ball 1 moving along ' + axis + ' axis', fontsize=fontsize)
ax[idx, 0].plot(x_ticks, curr_latent[:, 0], marker= marker_list[0], markersize= fontsize_lgd/1.2, linewidth=4, ls='-', label='Latent 1')
ax[idx, 0].plot(x_ticks, curr_latent[:, 1], marker= marker_list[1], markersize= fontsize_lgd/1.2, linewidth=4, ls='--', label='Latent 2')
ax[idx, 0].plot(x_ticks, curr_latent[:, 2], marker= marker_list[2], markersize= fontsize_lgd/1.2, linewidth=4, ls='-.', label='Latent 3')
ax[idx, 0].plot(x_ticks, curr_latent[:, 3], marker= marker_list[3], markersize= fontsize_lgd/1.2, linewidth=4, ls=':', label='Latent 4')
curr_latent= pred_latent[4, :, :]
ax[idx, 1].grid(True)
ax[idx, 1].set_ylim(-2.50, 2.50)
ax[idx, 1].tick_params(labelsize=fontsize)
# ax[idx, 1].set_ylabel('Predicted Latents', fontsize=fontsize)
ax[idx, 1].set_xlabel('Ball 2 moving along ' + axis + ' axis', fontsize=fontsize)
ax[idx, 1].plot(x_ticks, curr_latent[:, 0], marker= marker_list[0], markersize= fontsize_lgd/1.2, linewidth=4, ls='-', label='Latent 1')
ax[idx, 1].plot(x_ticks, curr_latent[:, 1], marker= marker_list[1], markersize= fontsize_lgd/1.2, linewidth=4, ls='--', label='Latent 2')
ax[idx, 1].plot(x_ticks, curr_latent[:, 2], marker= marker_list[2], markersize= fontsize_lgd/1.2, linewidth=4, ls='-.', label='Latent 3')
ax[idx, 1].plot(x_ticks, curr_latent[:, 3], marker= marker_list[3], markersize= fontsize_lgd/1.2, linewidth=4, ls=':', label='Latent 4')
lines, labels = fig.axes[-1].get_legend_handles_labels()
lgd= fig.legend(lines, labels, loc="lower center", bbox_to_anchor=(0.5, -0.08), fontsize=fontsize_lgd, ncol=4)
plt.tight_layout()
plt.savefig(save_dir + 'latent_traversal_seed_' + str(seed) + '.pdf', bbox_extra_artists=(lgd,), bbox_inches='tight')
plt.clf()
#Save the metrics dataframe containing results for the different random seeds
if target_seed == -1:
print('Dataframe')
print(res)
f= open(save_dir + 'logs.pickle', 'wb')
pickle.dump(res, f)
for key in res.keys():
res[key]= np.array(res[key])
print('Metric: ', key, np.mean(res[key]), np.std(res[key])/np.sqrt(num_seeds))