-
Notifications
You must be signed in to change notification settings - Fork 0
/
pipeline10.py
160 lines (141 loc) · 7.64 KB
/
pipeline10.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
import os
from time import localtime, strftime
import numpy as np
import matplotlib.pyplot as plt
from sklearn.impute import KNNImputer
from build.class_gain import GAIN
from build.class_misgan import MisGAN
from build.utils import normalization, load_dataset
np.random.seed(666)
LIST_DATASETS = ["breast", "credit", "letter", "news", "spam", "wine_red", "wine_white", "mydata1", "mydata2"]
NB_NEIGHBOURS = 50
NB_EPOCHS_GAIN = 20000
NB_EPOCHS_MISGAN = 5000
NB_REPEAT_TRAIN = 20
NB_REPEAT_IMPUTATION = 50
rmse_gain = np.zeros((NB_REPEAT_TRAIN, len(LIST_DATASETS)))
rmse_misgan = np.zeros((NB_REPEAT_TRAIN, len(LIST_DATASETS)))
rmse_knn1 = np.zeros((NB_REPEAT_TRAIN, len(LIST_DATASETS)))
rmse_knn2 = np.zeros((NB_REPEAT_TRAIN, len(LIST_DATASETS)))
for i1 in range(NB_REPEAT_TRAIN):
print(f"\n=================", flush=True)
print(f"== Repeat {i1} ==", end=" ")
print(strftime("%Y-%m-%d %H:%M:%S", localtime()), flush=True)
print(f"=================", flush=True)
for i2 in range(len(LIST_DATASETS)):
dataset_name = LIST_DATASETS[i2]
original_data = load_dataset(dataset_name)
original_data = np.array(original_data, dtype="float32")
mult_fact = original_data.shape[0] / 1000.0 # Multiplicative factor for epochs and nb_neighbours
print(f"-> Dataset = {dataset_name} {original_data.shape} | Time: ", end="", flush=True)
print(strftime("%Y-%m-%d %H:%M:%S", localtime()), flush=True)
r = np.random.uniform(size=original_data.shape)
MNARprob = np.zeros(shape=original_data.shape)
for c in range(original_data.shape[1]):
MNARprob[:, c] = np.argsort(np.argsort(original_data[:, c])) / original_data.shape[0] * 0.4
miss_mask = (r < MNARprob)
nb_miss_val = np.sum(miss_mask)
miss_data = np.copy(original_data)
miss_data[miss_mask] = np.nan
norm_data, norm_params = normalization(miss_data)
norm_full_data, _ = normalization(original_data, norm_params=norm_params)
myGAIN = GAIN(dim=original_data.shape[1])
myMisGAN = MisGAN(dim=original_data.shape[1])
cur_gain_epochs = int(NB_EPOCHS_GAIN / mult_fact)
cur_misgan_epochs = int(NB_EPOCHS_MISGAN / mult_fact)
print(f"Train GAIN for {cur_gain_epochs} epochs...", end=" ")
myGAIN.train(norm_data, batch_size=128, epochs=cur_gain_epochs)
print(f"Train MisGAN for {cur_misgan_epochs} epochs...", end=" ")
myMisGAN.train(norm_data, batch_size=128, epochs=cur_misgan_epochs)
impute_gain = np.zeros_like(miss_data)
impute_misgan = np.zeros_like(miss_data)
for i3 in range(NB_REPEAT_IMPUTATION):
impute_gain += myGAIN.impute(norm_data)
impute_misgan += myMisGAN.impute(norm_data)
impute_gain /= NB_REPEAT_IMPUTATION
impute_misgan /= NB_REPEAT_IMPUTATION
rmse_gain[i1, i2] = np.sqrt(np.sum((impute_gain - norm_full_data) ** 2) / nb_miss_val)
rmse_misgan[i1, i2] = np.sqrt(np.sum((impute_misgan - norm_full_data) ** 2) / nb_miss_val)
print("KNN starts ", end="") # For KNN
cur_knn_neighbours = int(NB_NEIGHBOURS * mult_fact)
print(f"({cur_knn_neighbours} neighbours...)")
print(strftime("%Y-%m-%d %H:%M:%S", localtime()))
myKNN1 = KNNImputer(n_neighbors=cur_knn_neighbours, weights="uniform", metric="nan_euclidean")
myKNN1.fit(norm_data)
impute_knn1 = myKNN1.transform(norm_data)
myKNN2 = KNNImputer(n_neighbors=cur_knn_neighbours, weights="distance", metric="nan_euclidean")
myKNN2.fit(norm_data)
impute_knn2 = myKNN2.transform(norm_data)
rmse_knn1[i1, i2] = np.sqrt(np.sum((impute_knn1 - norm_full_data) ** 2) / nb_miss_val)
rmse_knn2[i1, i2] = np.sqrt(np.sum((impute_knn2 - norm_full_data) ** 2) / nb_miss_val)
print("")
if not os.path.exists("results/pipeline10/"):
os.makedirs("results/pipeline10/")
if not os.path.exists("results/pipeline10/plots"):
os.makedirs("results/pipeline10/plots")
np.save("results/pipeline10/rmse_gain.npy", rmse_gain)
np.save("results/pipeline10/rmse_misgan.npy", rmse_misgan)
np.save("results/pipeline10/rmse_knn1.npy", rmse_knn1)
np.save("results/pipeline10/rmse_knn2.npy", rmse_knn2)
# Plot 1: Complete performances for everyone
mu1 = np.mean(rmse_gain, axis=0)
std1 = np.std(rmse_gain, axis=0)
mu2 = np.mean(rmse_misgan, axis=0)
std2 = np.std(rmse_misgan, axis=0)
mu3 = np.mean(rmse_knn1, axis=0)
std3 = np.std(rmse_knn1, axis=0)
mu4 = np.mean(rmse_knn2, axis=0)
std4 = np.std(rmse_knn2, axis=0)
x_ticks_locs = np.arange(0.0, 9.0)
plt.figure(figsize=(8, 4))
plt.errorbar(x_ticks_locs-0.15, mu1, yerr=std1, color="indigo", marker=".", capsize=5.0, ls="", label="GAIN")
plt.errorbar(x_ticks_locs-0.05, mu2, yerr=std2, color="grey", marker=".", capsize=5.0, ls="", label="MisGAN")
plt.errorbar(x_ticks_locs+0.05, mu3, yerr=std3, color="teal", marker=".", capsize=5.0, ls="", label="KNN-uniform")
plt.errorbar(x_ticks_locs+0.15, mu4, yerr=std4, color="peru", marker=".", capsize=5.0, ls="", label="KNN-distance")
plt.xticks(x_ticks_locs, LIST_DATASETS, rotation ="vertical")
plt.ylabel("RMSE")
plt.title("[MNAR setting] mean missing_rate of 20%")
plt.legend()
plt.tight_layout()
plt.savefig("results/pipeline10/plots/rmse.pdf")
# Plot 2: Remove the bad performances for the dataset News
mu1 = np.mean(rmse_gain, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std1 = np.std(rmse_gain, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
mu2 = np.mean(rmse_misgan, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std2 = np.std(rmse_misgan, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
mu3 = np.mean(rmse_knn1, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std3 = np.std(rmse_knn1, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
mu4 = np.mean(rmse_knn2, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std4 = np.std(rmse_knn2, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
x_ticks_locs = np.arange(0.0, 8.0)
x_ticks_names = ["breast", "credit", "letter", "spam", "wine_red", "wine_white", "mydata1", "mydata2"]
plt.figure(figsize=(8, 4))
plt.errorbar(x_ticks_locs-0.15, mu1, yerr=std1, color="indigo", marker=".", capsize=5.0, ls="", label="GAIN")
plt.errorbar(x_ticks_locs-0.05, mu2, yerr=std2, color="grey", marker=".", capsize=5.0, ls="", label="MisGAN")
plt.errorbar(x_ticks_locs+0.05, mu3, yerr=std3, color="teal", marker=".", capsize=5.0, ls="", label="KNN-uniform")
plt.errorbar(x_ticks_locs+0.15, mu4, yerr=std4, color="peru", marker=".", capsize=5.0, ls="", label="KNN-distance")
plt.xticks(x_ticks_locs, x_ticks_names, rotation ="vertical")
plt.ylabel("RMSE")
plt.title("[MNAR setting] mean missing_rate of 20%")
plt.legend()
plt.tight_layout()
plt.savefig("results/pipeline10/plots/rmse2.pdf")
# Plot 2: Remove the bad performances for the dataset News
mu1 = np.mean(rmse_gain, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std1 = np.std(rmse_gain, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
mu3 = np.mean(rmse_knn1, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std3 = np.std(rmse_knn1, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
mu4 = np.mean(rmse_knn2, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
std4 = np.std(rmse_knn2, axis=0)[[0, 1, 2, 4, 5, 6, 7, 8]]
x_ticks_locs = np.arange(0.0, 8.0)
x_ticks_names = ["breast", "credit", "letter", "spam", "wine_red", "wine_white", "mydata1", "mydata2"]
plt.figure(figsize=(8, 4))
plt.errorbar(x_ticks_locs-0.15, mu1, yerr=std1, color="indigo", marker=".", capsize=5.0, ls="", label="GAIN")
plt.errorbar(x_ticks_locs, mu3, yerr=std3, color="teal", marker=".", capsize=5.0, ls="", label="KNN-uniform")
plt.errorbar(x_ticks_locs+0.15, mu4, yerr=std4, color="peru", marker=".", capsize=5.0, ls="", label="KNN-distance")
plt.xticks(x_ticks_locs, x_ticks_names, rotation ="vertical")
plt.ylabel("RMSE")
plt.title("[MNAR setting] mean missing_rate of 20%")
plt.legend()
plt.tight_layout()
plt.savefig("results/pipeline10/plots/rmse3.pdf")