-
Notifications
You must be signed in to change notification settings - Fork 3
/
plot_sentence_stats.py
202 lines (160 loc) · 5.68 KB
/
plot_sentence_stats.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
import argparse
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from train_image_captioning import UNIQUE_VERBS
from utils import LEGEND_GROUPED_NOUNS, LEGEND
import numpy as np
LEGEND_STATS_SENTENCE_LENGTH = {
"seq_lengths": "sentence_length",
}
LEGEND_STATS_PERSONS = {
"jenny_occurrences": "jenny",
"mike_occurrences": "mike",
}
LEGEND_STATS_VERBS = {verb: verb for verb in UNIQUE_VERBS}
LEGEND_STATS_ALL = {
**LEGEND_STATS_SENTENCE_LENGTH,
**LEGEND_STATS_PERSONS,
**LEGEND_STATS_VERBS,
}
def main(args):
# sns.set_context(
# "paper",
# rc={
# "font.size": 12,
# "axes.titlesize": 12,
# "axes.labelsize": 12,
# "xtick.labelsize": 12,
# "ytick.labelsize": 12,
# "legend.fontsize": 12,
# },
# )
all_scores = []
for run, scores_file in enumerate(args.scores_files):
scores = pd.read_csv(scores_file)
for column_name in scores.columns:
if not (column_name == "epoch" or column_name == "batch_id"):
scores[column_name] = (
scores[column_name]
.rolling(args.rolling_window, min_periods=1)
.mean()
)
scores.set_index("num_samples", inplace=True)
scores.rename(columns=LEGEND_STATS_ALL, inplace=True)
metric = (
"bleu_score_val"
if "bleu_score_val" in scores.columns
else "bleu_score_train"
)
best_score = scores[scores[metric] == scores[metric].max()]
if len(best_score) == 0:
print("No best score, taking last value!")
best_score = scores.tail(1)
if "epoch" in scores.columns:
print(f"Epoch with max {metric}:{best_score['epoch'].values[0]}")
print(f"num_samples with max {metric}:{best_score.index.values[0]}")
# overall_average = np.mean([best_score[name].values[0] for name in LEGEND.values()])
# print(f"Overview Average: {overall_average:.3f}")
# Read train setup information from file name
setup = str(scores_file.split("/")[7])
scores_setup = []
for row in scores.iterrows():
for name in LEGEND_STATS_ALL.values():
filtered_scores = {}
filtered_scores["score"] = name
filtered_scores["value"] = row[1][name]
filtered_scores["num_samples"] = row[0]
filtered_scores["setup"] = setup
scores_setup.append(filtered_scores)
all_scores.extend(scores_setup.copy())
fig, axes = plt.subplots(1, 2, sharey="row", sharex="all")
all_scores = pd.DataFrame(all_scores)
for axis_y, setup in enumerate(all_scores.setup.unique()):
legend = LEGEND_STATS_PERSONS
legend_values = legend.values()
scores_setup = all_scores[(all_scores.setup == setup) & (all_scores.score.isin(legend_values))]
legend = True if axis_y == 1 else False
g = sns.lineplot(
ax=axes[axis_y],
data=scores_setup,
x="num_samples",
y="value",
hue="score",
legend=legend,
)
axes[axis_y].set_ylabel("Occurrences/Sentence")
if axis_y == 1:
g.legend(loc='best', fontsize=9, ncol=2)
axes[0].set_title("XSL+CF")
axes[1].set_title("XSL+Alt")
if args.x_lim:
plt.xlim((0, args.x_lim))
plt.tight_layout()
plt.show()
fig, axes = plt.subplots(1, 2, sharey="row", sharex="all") # figsize=(15, 5)
for axis_y, setup in enumerate(all_scores.setup.unique()):
legend = LEGEND_STATS_VERBS
legend_values = legend.values()
scores_setup = all_scores[(all_scores.setup == setup) & (all_scores.score.isin(legend_values))]
legend = True if axis_y == 1 else False
g = sns.lineplot(
ax=axes[axis_y],
data=scores_setup,
x="num_samples",
y="value",
hue="score",
legend=legend,
)
axes[axis_y].set_ylabel("Occurrences/Sentence")
if axis_y == 1:
g.legend(loc='best', fontsize=9, ncol=2)
axes[0].set_title("XSL+CF")
axes[1].set_title("XSL+Alt")
if args.x_lim:
plt.xlim((0, args.x_lim))
plt.tight_layout()
plt.show()
fig, axes = plt.subplots(1, 2, sharey="row", sharex="row") # figsize=(15, 5)
all_scores = pd.DataFrame(all_scores)
for axis_y, setup in enumerate(all_scores.setup.unique()):
legend = LEGEND_STATS_SENTENCE_LENGTH
legend_values = legend.values()
scores_setup = all_scores[(all_scores.setup == setup) & (all_scores.score.isin(legend_values))]
sns.lineplot(
ax=axes[axis_y],
data=scores_setup,
x="num_samples",
y="value",
hue="score",
# style="score",
legend=False,
)
axes[axis_y].set_ylabel("Mean Sentence Length")
axes[0].set_title("XSL+CF")
axes[1].set_title("XSL+Alt")
if args.x_lim:
plt.xlim((0, args.x_lim))
plt.tight_layout()
plt.show()
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument(
"--scores-files", type=str, nargs="+", required=True,
)
parser.add_argument(
"--rolling-window", default=1, type=int,
)
parser.add_argument(
"--x-lim", default=None, type=int,
)
parser.add_argument(
"--y-lim", default=1.0, type=float,
)
parser.add_argument(
"--group-noun-accuracies", default=False, action="store_true",
)
return parser.parse_args()
if __name__ == "__main__":
args = get_args()
main(args)