-
Notifications
You must be signed in to change notification settings - Fork 0
/
analyze_multimodal.py
77 lines (57 loc) · 2.44 KB
/
analyze_multimodal.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
"""
Plot results for experiments on multimodal data
"""
import numpy as np
from sklearn.metrics import f1_score
from utils import multilabel_weighted_accuracy
import matplotlib.pyplot as plt
from tqdm import tqdm
import matplotlib
matplotlib.rcParams.update({'font.size': 16, "font.family" : "monospace"})
n_folds = 10
# IMG|TXT|LF x METRICS x FOLDS
scores = np.zeros((3, 3, 10))
metrics = ["$F_1$ micro", "$F_1$ macro", "MWA"]
colors = ["tomato", "dodgerblue", "limegreen"]
modalities = ["IMG", "LF", "TXT"]
# IMG losses
losses = np.load("preds_img/epochs_loss_r18.npy")
for fold_id in tqdm(range(n_folds)):
txt_probas = np.load("preds/exp_2/preds_%i.npy" % fold_id)
txt_y = np.load("preds/exp_2/test_%i.npy" % fold_id)
img_probas = np.load("preds_img/fold_%i_probas_r18.npy" % fold_id)
img_y = np.load("preds_img/fold_%i_test_r18.npy" % fold_id)
if not np.array_equal(txt_y, img_y):
print("Wrong labels!")
img_preds = (img_probas > .5).astype(int)
txt_preds = (txt_probas > .5).astype(int)
lf_preds = ((img_probas+txt_probas)/2 > .5).astype(int)
all_preds = [img_preds, lf_preds, txt_preds]
for preds_id, preds in enumerate(all_preds):
f1_micro = f1_score(img_y, preds, average='micro')
f1_macro = f1_score(img_y, preds, average='macro')
mwa = multilabel_weighted_accuracy(img_y, preds)
scores[preds_id, 0, fold_id] = f1_micro
scores[preds_id, 1, fold_id] = f1_macro
scores[preds_id, 2, fold_id] = mwa
fig, ax = plt.subplots(1, 1, figsize=(15, 6))
mean_scores= np.mean(scores, axis=2)
std_scores= np.std(scores, axis=2)
x = np.arange(len(metrics)) # the label locations
width = 0.25 # the width of the bars
multiplier = 0
for score_id in range(3):
offset = width * multiplier
rects = ax.bar(x + offset, np.round(mean_scores, 3)[score_id], width, label=modalities[score_id], color=colors[score_id])
ax.bar_label(rects, padding=25)
ax.errorbar(x + offset, np.round(mean_scores, 3)[score_id], std_scores[score_id], fmt='.', color='Black', elinewidth=10,capthick=30,errorevery=1, alpha=None, ms=4, capsize = 2)
multiplier += 1
ax.set_xticks(x + width, metrics)
ax.legend(frameon=False)
ax.grid(ls=":", c=(0.7, 0.7, 0.7))
ax.set_ylabel("score")
ax.set_ylim(0, 0.5)
ax.spines[['right', 'top']].set_visible(False)
plt.tight_layout()
plt.savefig("figures/multimodal.png", dpi=200)
plt.savefig("figures/multimodal.eps", dpi=200)