-
Notifications
You must be signed in to change notification settings - Fork 9
/
create_plots.py
296 lines (252 loc) · 10.4 KB
/
create_plots.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
import argparse
import pandas
import os
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
def get_category_order_labels(category_order, attribute, model, row):
""" Hand-crafted category order and corresponding labels for x-axis in box plots """
labels = []
if attribute == "train_type_loss":
category_order = ['negative_sampling-margin_ranking',
'negative_sampling-bce',
'1vsAll-bce',
'KvsAll-bce',
'negative_sampling-kl',
'1vsAll-kl',
'KvsAll-kl'
]
if not len(labels):
labels = ['NegSamp+MR',
'NegSamp+BCE', '1vsAll+BCE', 'KvsAll+BCE',
'NegSamp+CE', '1vsAll+CE', 'KvsAll+CE']
elif attribute == "reciprocal":
if current_model == "conve":
category_order = [1]
labels = ["Reciprocal"]
else:
category_order = [0, 1]
labels = ["No Reciprocal", "Reciprocal"]
elif attribute == "dropout_e":
category_order = [0, 1]
labels = ["No Dropout", "Dropout"]
elif attribute == "dropout_r":
category_order = [0, 1]
labels = ["No Dropout", "Dropout"]
elif attribute == "dropout":
category_order = [0, 1]
labels = ["No Dropout", "Dropout"]
elif attribute == "emb_initialize":
category_order = ["normal_", "uniform_", "xavier_normal_", "xavier_uniform_"]
labels = ["Normal", "Unif.", "XvNorm", "XvUnif"]
elif attribute == "emb_regularize_p":
category_order = [0.0, 1.0, 2.0, 3.0]
labels = ["None", "L1", "L2", "L3"]
else:
labels = category_order
# no labels for top row
if row == 0:
labels = []
for n in range(len(category_order)):
labels.append('')
return category_order, labels
if __name__ == '__main__':
# parse args
parser = argparse.ArgumentParser()
parser.add_argument('--csv', type=str, required=True, help="csvs created with dump command, comma separated")
parser.add_argument('--output_folder', type=str, required=True, help="name of output folder")
args, _ = parser.parse_known_args()
# determine which attributes to plot
attributes = ['train_batch_size',
'train_type_loss',
'train_optimizer',
'emb_e_dim',
'emb_regularize_p',
'emb_initialize',
'reciprocal',
'dropout'
]
# load input CSVs
csvs = []
for input_file in args.csv.split(","):
csvs.append(pandas.read_csv(input_file))
all_data = pandas.concat(csvs)
# add train type and loss combination column
all_data['train_type_loss'] = all_data['train_type'] + '-' + all_data['train_loss']
all_data = all_data.drop(columns=["train_type", "train_loss"])
# add dropout/no dropout column
all_data['dropout_e'] = np.where(all_data.emb_e_dropout > 0, 1, 0)
all_data['dropout_r'] = np.where(all_data.emb_r_dropout > 0, 1, 0)
all_data['dropout'] = np.where((all_data.emb_e_dropout > 0) | (all_data.emb_r_dropout > 0), 1, 0)
# deal with empty string in emb_regularize_p
all_data['emb_regularize_p'] = all_data['emb_regularize_p'].fillna(0)
# make sure no Bayesian trials are included
all_data = all_data.loc[~all_data['folder'].str.contains("-bo")]
# create output folder
output_folder = args.output_folder
if os.path.isdir(output_folder):
raise ValueError('Output folder already exists.')
os.mkdir(output_folder)
# set datasets
datasets = ["fb15k-237", "wnrr"]
dataset_labels = ["FB15K-237", "WNRR"]
# set models
models = all_data.model.unique()
# set metric label
metric_label = "Validation MRR"
# order and label for models
model_order = ['rescal', 'transe', 'distmult', 'complex', 'conve']
model_labels = ['RESCAL', 'TransE', 'DistMult', 'ComplEx', 'ConvE']
# plot MRR per model
print("Plotting metric per model...")
f, axes = plt.subplots(1, len(datasets), sharex=True, sharey=True)
f.set_size_inches(6.4, 2.4)
plt.subplots_adjust(hspace=0.2, wspace=0.1)
font_size = 7.5
label_rotation=15
for i in range(len(datasets)):
current_dataset = datasets[i]
title = dataset_labels[i]
# get data for current dataset
dataset_data = all_data.loc[all_data['dataset'] == current_dataset]
# create plots
mrr_per_model = sns.boxplot(y=dataset_data['metric']*100,
x=dataset_data['model'],
order=model_order,
linewidth=0.5,
fliersize=1,
width=0.8,
ax=axes[i])
mrr_per_model.set_title(title, size=font_size)
mrr_per_model.tick_params(labelsize=font_size)
if i == 0:
mrr_per_model.set_ylabel(metric_label, fontsize=font_size)
else:
mrr_per_model.set_ylabel('')
for ax in axes.flat:
ax.set_xticklabels(model_labels)
ax.set(xlabel='')
# save figure
mrr_per_model.get_figure().savefig(
output_folder + '/metric_per_model.pdf',
dpi=300,
bbox_inches='tight',
pad_inches=0)
# plot each attribute vs metric
for attribute in attributes:
attribute_name = attribute
print("Plotting {}...".format(attribute_name))
# create plot
label_rotation=30
font_size = 7.5
num_rows = len(datasets)
num_cols = len(models)
# manage with/no penalty dropouts
if "_no_penalty" in attribute_name:
attribute = attribute[:-len("_no_penalty")]
elif "_with_penalty" in attribute_name:
attribute = attribute[:-len("_with_penalty")]
# skip Conve if reciprocal
if attribute == "reciprocal":
num_cols = len(models) - 1
f, axes = plt.subplots(num_rows, num_cols, sharex=True, sharey=True)
if attribute == "train_type_loss":
f.set_size_inches(6.1, 3.5)
elif attribute == "reciprocal":
f.set_size_inches(4.5, 4)
else:
f.set_size_inches(6, 4)
plt.subplots_adjust(hspace=0.1, wspace=0.2)
plt.xticks(rotation=label_rotation)
# add dataset names to subplot rows
row_names = ["FB15K-237", "WNRR"]
pad = 5
for ax, row in zip(axes[:,0], row_names):
ax.annotate(row, xy=(0, 0.5), rotation=90, xytext=(-ax.yaxis.labelpad - pad, 0),
xycoords=ax.yaxis.label, textcoords='offset points',
size=font_size, ha='right', va='center')
# each model is a column in the main plot
for col in range(len(model_order)):
current_model = model_order[col]
title = model_labels[col]
# skip Conve if reciprocal
if attribute == "reciprocal" and current_model == "conve":
continue
# get data for current model
model_data = all_data.loc[all_data['model'] == current_model]
# manage with/no penalty dropouts
if "_no_penalty" in attribute_name:
model_data = model_data.loc[model_data['emb_regularize_p'] == 0]
elif "_with_penalty" in attribute_name:
model_data = model_data.loc[model_data['emb_regularize_p'] > 0]
# each dataset is a row in the main plot
for row in range(len(datasets)):
current_dataset = datasets[row]
# set order and labels for categories in x axis of box plot
category_order = sorted(model_data[attribute].fillna(0).unique())
category_order, labels = get_category_order_labels(category_order, attribute, current_model, row)
if attribute == "train_type_loss":
global_legends = labels
labels = ['', '', '', '', '', '', '']
dataset_data = model_data.loc[model_data['dataset'] == current_dataset]
if attribute == "train_type_loss":
colors=sns.color_palette()
box = sns.boxplot(x=dataset_data[attribute],
y=dataset_data['metric']*100,
order=category_order,
linewidth=0.5,
fliersize=1,
palette=colors,
ax=axes[row][col]
)
else:
box = sns.boxplot(x=dataset_data[attribute],
y=dataset_data['metric']*100,
order=category_order,
linewidth=0.5,
fliersize=1,
ax=axes[row][col]
)
if row != len(datasets) - 1:
box.set_title(title, size=font_size)
box.tick_params(labelsize=font_size)
if col == 0:
box.set_ylabel(metric_label, fontsize=font_size)
else:
box.set_ylabel('')
# add xticks labels
axes[row][col].set_xticklabels(labels, ha='right')
# add labels to box plots
for ax in axes.flat:
plt.sca(ax)
plt.xticks(rotation=label_rotation)
ax.set(xlabel='')
# add legend
if attribute == "train_type_loss":
proxies = []
for i in range(len(global_legends)):
proxies.append(plt.Rectangle(
(0,0),
1,
1,
ec='k',
fc=colors[i],
linewidth=0.5,
label=global_legends[i]))
f.legend(
prop={'size': 5.5},
handles=proxies,
bbox_to_anchor=(0.01, 0.02),
loc='lower left',
borderaxespad=0.,
columnspacing=1,
ncol=7,
frameon=False,
)
# save figure for current attribute
box.get_figure().savefig(
output_folder + '/' + attribute_name + '.pdf',
dpi=300,
bbox_inches='tight',
pad_inches=0)
print('DONE! Find plots in folder:', output_folder)