forked from VincentGranville/Main
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathNoGAN_library_sudents.py
200 lines (160 loc) · 7.44 KB
/
NoGAN_library_sudents.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
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import matplotlib as mpl
import genai_evaluation as ge
from genai_evaluation import multivariate_ecdf, ks_statistic
import nogan_synthesizer as ns
from nogan_synthesizer import NoGANSynth
from nogan_synthesizer.preprocessing import wrap_category_columns, unwrap_category_columns
import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)
pd.core.common.random_state(None)
seed = 1003
np.random.seed(seed)
#--- [1] read data, feature selection, remove missing values
url = "https://raw.githubusercontent.com/VincentGranville/Main/main/students.csv"
data = pd.read_csv(url)
data = data.drop(data[data["Curricular_units_2nd_sem_grade"] == 0].index)
features = [
'Curricular_units_2nd_sem_approved',
'Curricular_units_2nd_sem_grade',
'Curricular_units_1st_sem_approved',
'Curricular_units_1st_sem_grade',
'Admission_grade',
'Tuition_fees_up_to_date',
'Curricular_units_2nd_sem_evaluations',
'Age_at_enrollment',
'Previous_qualification_grade',
'Curricular_units_1st_sem_evaluations',
'Course',
'Father_occupation',
'Mother_occupation',
'Unemployment_rate',
'GDP',
'Application_mode',
'Father_qualification',
'Curricular_units_2nd_sem_enrolled',
'Mother_qualification',
'Inflation_rate',
'Target']
target_column = 'Target'
cat_cols = ['Course', 'Father_occupation', 'Mother_occupation',
'Unemployment_rate', 'GDP', 'Inflation_rate',
'Application_mode', 'Father_qualification',
'Mother_qualification', 'Tuition_fees_up_to_date', 'Target'
]
num_cols = [f for f in features if f not in cat_cols]
data = data[features]
#--- [2] Split real data into training and validation sets
training_data = data.sample(frac = 0.5)
validation_data = data.drop(training_data.index)
#--- [3] Preprocess Categorical Columns
wrapped_train_data, idx_to_key_train, key_to_idx_train = \
wrap_category_columns(training_data,cat_cols)
wrapped_val_data, idx_to_key_val, key_to_idx_val = \
wrap_category_columns(validation_data,cat_cols)
#--- [4] Train model
bin_values = [100] * len(wrapped_train_data.columns)
nogan = NoGANSynth(wrapped_train_data,random_seed=seed)
nogan.fit(bins = bin_values)
#--- [5] Generate 3 synth. data: original NoGAN, Gaussian, Uniform
stretch_type_gaussian = (["Gaussian"] * (len(wrapped_train_data.columns)-1)) + ["Uniform"]
stretch_nogan_orig = [-1] * len(wrapped_train_data.columns)
print("Synth. NoGAN Original:")
wrapped_nogan_orig_synth_data = nogan.generate_synthetic_data(len(wrapped_train_data),
debug = True ,
stretch = stretch_nogan_orig
)
print("\nSynth. NoGAN Gaussian:")
wrapped_nogan_gauss_synth_data = nogan.generate_synthetic_data(len(wrapped_train_data),
debug = True ,
stretch_type= stretch_type_gaussian
)
print("\nSynth. NoGAN Uniform:")
wrapped_nogan_uniform_synth_data = nogan.generate_synthetic_data(len(wrapped_train_data),
debug = True ,
)
print()
#--- [6] Evaluate using ECDF & KS Stat
_, ecdf_val1, ecdf_nogan_orig_synth = \
multivariate_ecdf(wrapped_val_data,
wrapped_nogan_orig_synth_data,
n_nodes = 1000,
verbose = True,
random_seed=seed)
_, ecdf_val2, ecdf_nogan_gauss_synth = \
multivariate_ecdf(wrapped_val_data,
wrapped_nogan_gauss_synth_data,
n_nodes = 1000,
verbose = True,
random_seed=seed)
_, ecdf_val3, ecdf_nogan_uniform_synth = \
multivariate_ecdf(wrapped_val_data,
wrapped_nogan_uniform_synth_data,
n_nodes = 1000,
verbose = True,
random_seed=seed)
_, ecdf_val4, ecdf_train = \
multivariate_ecdf(wrapped_val_data,
wrapped_train_data,
n_nodes = 1000,
verbose = True,
random_seed=seed)
print()
ks_stat_nogan_orig = ks_statistic(ecdf_val1, ecdf_nogan_orig_synth)
ks_stat_nogan_gauss = ks_statistic(ecdf_val2, ecdf_nogan_gauss_synth)
ks_stat_nogan_uniform = ks_statistic(ecdf_val3, ecdf_nogan_uniform_synth)
ks_stat_train = ks_statistic(ecdf_val4, ecdf_train)
print(f"KS Stat NoGAN Original(Synth vs Validation): {ks_stat_nogan_orig:.5f}")
print(f"KS Stat NoGAN Gauss (Synth vs Validation): {ks_stat_nogan_gauss:.5f}")
print(f"KS Stat NoGAN Uniform (Synth vs Validation): {ks_stat_nogan_uniform:.5f}")
print(f"Base KS Stat (Synth vs Train): {ks_stat_train:.5f}")
#--- [7] Expand Categorical Columns in Synth Data
nogan_orig_synth_data = unwrap_category_columns(data=wrapped_nogan_orig_synth_data,
idx_to_key=idx_to_key_train, cat_cols=cat_cols)
nogan_orig_synth_data = nogan_orig_synth_data[features]
nogan_gauss_synth_data = unwrap_category_columns(data=wrapped_nogan_gauss_synth_data,
idx_to_key=idx_to_key_train, cat_cols=cat_cols)
nogan_gauss_synth_data = nogan_gauss_synth_data[features]
nogan_uniform_synth_data = unwrap_category_columns(data=wrapped_nogan_uniform_synth_data,
idx_to_key=idx_to_key_train, cat_cols=cat_cols)
nogan_uniform_synth_data = nogan_uniform_synth_data[features]
#--- [8] Evaluation scatterplots
dfs_orig = nogan_orig_synth_data
dfs_gauss = nogan_gauss_synth_data
dfs_uniform = nogan_uniform_synth_data
dfv = validation_data
def vg_scatter(df, feature1, feature2, counter, xlim, ylim):
# customized plots, subplot position based on counter
label = feature1[0:20] + " vs " + feature2[0:20]
df = df[(df[feature1] >= xlim[0]) &
(df[feature1] <= xlim[1]) &
(df[feature2] >= ylim[0]) &
(df[feature2] <= ylim[1])
]
x = df[feature1].to_numpy()
y = df[feature2].to_numpy()
plt.subplot(3, 2, counter)
plt.scatter(x, y, s = 0.1, c ="blue")
plt.xlabel(label, fontsize = 7)
plt.xticks([])
plt.yticks([])
return()
mpl.rcParams['axes.linewidth'] = 0.3
[col1, col2] = [features[4], features[7]]
xlim = [min(dfv[col1]), max(dfv[col1])]
ylim = [min(dfv[col2]), max(dfv[col2])]
vg_scatter(dfs_uniform, col1, col2, 1, xlim, ylim)
vg_scatter(dfv, col1, col2, 2, xlim, ylim)
[col1, col2] = [features[4], features[3]]
xlim = [min(dfv[col1]), max(dfv[col1])]
ylim = [10, max(dfv[features[3]])]
vg_scatter(dfs_uniform, features[4], features[3], 3, xlim, ylim)
vg_scatter(dfv, features[4], features[3], 4, xlim, ylim)
[col1, col2] = [features[2], features[0]]
xlim = [min(dfv[features[2]]), max(dfv[features[0]])]
ylim = [min(dfv[features[2]]), max(dfv[features[0]])]
vg_scatter(dfs_uniform, features[2], features[0], 5, xlim, ylim)
vg_scatter(dfv, features[2], features[0], 6, xlim, ylim)
plt.show()