forked from szeighami/NeuroComplete
-
Notifications
You must be signed in to change notification settings - Fork 0
/
plt_embedding_dist.py
168 lines (129 loc) · 5.63 KB
/
plt_embedding_dist.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
from sklearn.manifold import TSNE
import numpy as np
import matplotlib
from sklearn.decomposition import PCA
from sklearn.metrics.pairwise import cosine_similarity
from sklearn.metrics import pairwise_distances
#matplotlib.rcParams['text.usetex'] = True
matplotlib.rcParams.update({'font.size': 24})
matplotlib.rcParams.update({'legend.frameon': False})
#matplotlib.rcParams.update({'font.sans-serif': ['Helvetica']})
matplotlib.rcParams.update({'font.family': 'serif'})
matplotlib.rcParams.update({'text.usetex': False})
matplotlib.rcParams.update({'lines.markersize': 12})
import matplotlib.pyplot as plt
alphabet = {0:'a', 1:'b', 2:'c'}
setting_name = ["H1", "M1"]
def plot(data_list, pre_pca, alphas, labels, title, ax):
all_samples = np.concatenate(data_list, axis=0)
if pre_pca:
all_samples = PCA(n_components=50).fit_transform(all_samples)
res = TSNE(n_components=2, learning_rate='auto',init='pca').fit_transform(all_samples)
curr_begin= 0
for i in range(len(data_list)):
curr_end = curr_begin+data_list[i].shape[0]
print(curr_begin, curr_end, curr_end-curr_begin)
if i == len(data_list)-1:
ax.plot(res[curr_begin:curr_end, 0], res[curr_begin:curr_end, 1], 'o', alpha=alphas[i], label=labels[i], c='k')
else:
ax.plot(res[curr_begin:curr_end, 0], res[curr_begin:curr_end, 1], 'o', alpha=alphas[i], label=labels[i])
curr_begin = curr_end
ax.set_xlabel("Dim. 1")
ax.set_ylabel("Dim. 2")
ax.set_title(title)
#ax.savefig(name)
def get_test_train(setting, sel, bias):
path = "tests/bias_factor_"+str(bias)+"/"+setting+"_selperc"+str(sel)+"_biased/"
test = np.load(path+'test_queries.npy')
test_res = np.load(path+'test_res.npy').astype(float).reshape((-1, 1))
train = np.load(path+'queries.npy')
res = np.load(path+'res.npy').astype(float).reshape((-1, 1))
return test, train
#settings = ["H1_AVG","H2_AVG","M1_AVG", "M2_AVG"]
settings = ["H1_AVG","M1_AVG"]
fig, axes = plt.subplots(nrows=1, ncols=3,figsize=(16.8, 5.3))
for i, setting in enumerate(settings):
test_best, train_best = get_test_train(setting, 0.8, 0.6)
test_worst, train_worst = get_test_train(setting, 0.05, 1.0)
selector = np.concatenate([train_best, train_worst], axis=0)
std_thresh = 0.01
selector = (selector - np.min(selector, axis=0))/(np.max(selector, axis=0)-np.min(selector, axis=0)+1e-5)
selector_vals = np.std(selector, axis=0)
test_best = test_best[:,np.where(selector_vals > std_thresh)[0]]
train_best = train_best[:,np.where(selector_vals > std_thresh)[0]]
train_worst = train_worst[:,np.where(selector_vals > std_thresh)[0]]
test_true = test_best[0::3]
ax = axes[i]
plot([train_best, train_worst, test_true], False, [0.2, 0.2, 1], ["Train (low bias)", "Train (high bias)", "Test"], f"({alphabet[i]}) {setting_name[i]} embedding", ax)
leg = ax.legend(loc='upper center', ncol=4, bbox_to_anchor=(-0.3, 1.35), fontsize=24, labelspacing=0.1, columnspacing=1, handletextpad=0.1)
for lh in leg.legendHandles:
lh.set_alpha(1)
sels = [0.05,0.8]
biases = [0.6,0.8, 1.0]
settings = ["H1_AVG","H2_AVG","M1_AVG", "M2_AVG"]
#settings = ["M2_AVG"]
bar_width = 0.25
low = []
high = []
for setting in settings:
data_list = []
for sel in sels:
for bias in biases:
path = "tests/bias_factor_"+str(bias)+"/"+setting+"_selperc"+str(sel)+"_biased/"
train = np.load(path+'queries.npy')
data_list.append(train)
test_best, _ = get_test_train(setting, 0.8, 0.6)
test_true = test_best[0::3]
data_list.append(test_true)
all_data = np.concatenate(data_list, axis=0)
selector = all_data
selector = (selector - np.min(selector, axis=0))/(np.max(selector, axis=0)-np.min(selector, axis=0)+1e-5)
selector_vals = np.std(selector, axis=0)
std_thresh =0.01
mask = np.where(selector_vals > std_thresh)[0]
all_data = all_data[:, mask]
test_true = all_data[-len(test_true):]
i = 0
curr_begin = 0
for sel in sels:
for bias in biases:
curr_end = curr_begin+len(data_list[i])
curr_train = all_data[curr_begin:curr_end]
curr_begin=curr_end
i += 1
dist = pairwise_distances(curr_train, test_true)
sim = cosine_similarity(curr_train, test_true)
#print(sim)
max_sim = np.max(sim, axis=0)
min_dist = np.min(dist, axis=0)
#print(min_sim.shape)
#print(min_sim)
avg_sim =np.mean(max_sim)
avg_dist =np.mean(min_dist)
#print(setting, sel, bias, avg_sim, avg_dist)
if sel == 0.05 and bias == 1.0:
low.append(avg_dist)
if sel == 0.8 and bias == 0.6:
high.append(avg_dist)
br1 = np.arange(len(settings))
br2 = [x + bar_width for x in br1]
ax=axes[2]
ax.bar(br2, high, width = bar_width, label ='Low bias')
ax.bar(br1, low, width = bar_width, label ='High bias')
ax.set_xlabel('Setting')
ax.set_ylabel('Dist. NTS')
ax.set_yscale('log')
ax.set_xticks([r+ bar_width/2 for r in range(len(settings))], ['H1', 'H2', 'M1', "M2"])
ax.legend(loc='upper center', ncol=4, bbox_to_anchor=(0.45, 1.35), fontsize=24, labelspacing=0.1, columnspacing=1, handletextpad=0.25)
ax.set_title(f"({alphabet[2]}) Embedding distance")
plt.subplots_adjust(left=0.08, bottom=0.17, right=0.98, top=0.82, wspace=0.29)
box = axes[1].get_position()
box.x0 = box.x0 - 0.005
box.x1 = box.x1 - 0.005
axes[1].set_position(box)
box = axes[0].get_position()
box.x0 = box.x0 - 0.007
box.x1 = box.x1 - 0.007
axes[0].set_position(box)
plt.savefig("embeddin_dist.png")
#plt.savefig("embeddin_dist.eps")