-
Notifications
You must be signed in to change notification settings - Fork 1
/
3_ABaugment.py
executable file
·225 lines (194 loc) · 9.75 KB
/
3_ABaugment.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
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
"""this script reproduce Table 2 of the paper:
Not as simple as we thought: a rigorous examination of data aggregation in materials informatics
09 August 2023, Version 1
Federico Ottomano, Giovanni De Felice, Vladimir Gusev, Taylor Sparks """
import numpy as np
import pandas as pd
import assets.plots as plots
from assets import utils
from assets import tasks
import pickle
from assets.preprocessing import preprocess_dataset, add_column
from settings import *
from aggr_models.baseline import elem_concat, concat
from aggr_models.discover_augmentation_v2 import DiscoAugment
from settings import *
import warnings
warnings.filterwarnings('ignore')
"""PROPERTIES"""
props_list = [
# 'thermalcond',
# 'seebeck',
# 'rho',
# 'sigma',
# 'bandgap'
'bulkmodulus',
'shearmodulus'
] # 'thermalcond',
# 'superconT',
# 'seebeck',
# 'rho'
# 'sigma',
# 'bandgap',
# 'bulkmodulus',
# 'shearmodulus'
"""TASKS"""
tasks_list = [
# # 'roost_regression'
# 'crabnet_regression'd,
# 'linear_regression',
'transf_crabnet_regression',
'transf_roost_regression'
# 'random_forest_regression',
# 'logistic_classification',
# 'crabnet_classification'
]
"""MODELS"""
models_list = [
# 'baseline',
'transfer'
# 'concat',
# 'elem_concat',
# 'disco'
]
"""SETTINGS""" #override from settings.py
k_elemconcat = 5 # n elements
n_elemconcat = 10 # n points per element
# metrics
metric_reg = 'mae'
metric_class = 'acc'
def plot_all():
# To store results
iterables = [[task for task in tasks_list], [model for model in models_list]]
columns = pd.MultiIndex.from_product(iterables, names=["task", "model"])
iterables = [props_list, list(range(n_repetitions))]
index = pd.MultiIndex.from_product(iterables, names=["prop", "rep"])
results = pd.DataFrame(columns=columns, index=index)
# main loop
for prop in props_list:
"""LOADING"""
# load datsets
data_raw = utils.load_dataset(prop)
keys_list = list(data_raw.keys())
key_A = pairs[prop][0]; assert key_A != 'mpds'
key_B = pairs[prop][1]
utils.print_info(data_raw, prop); print('')
"""PREPROCESSING"""
# preprocessing
data_clean = preprocess_dataset(data_raw, prop, merging,
epsilon_T,
med_sigma_multiplier,
mult_outliers,
ascending_setting[prop])
print(''); utils.print_info(data_clean, prop)
# add extraord column to all datasets(0/1)
data_clean = add_column(data_clean, extraord_size, ascending_setting[prop])
# ITERATE over different random seed to estimate unc.
for n in range(n_repetitions):
print(f'### seed = {n+1} ###')
seed = n
# SPLIT DATASETS IN TRAIN AND TEST
train, _, test = tasks.apply_split(split_type = split,
df = data_clean[key_A],
val_size=0, test_size=test_size, k_test=0.5,
random_state=seed,
ascending=ascending_setting[prop],
shuffle=shuffle_after_split)
data_B = data_clean[key_B]
# FEATURIZE TEST
test_feat = utils.featurize(test, elem_prop=elem_prop)
"""BASELINE"""
# does not merge datasets, every dataset is tested on its own test
if 'baseline' in models_list:
print(f'--- baseline ---')
# FEATURIZE TRAIN
train_feat = utils.featurize(train, elem_prop=elem_prop)
# TASKS
output, _ = tasks.apply_all_tasks(train_feat, test_feat, key_A,
tasks_list, prop, key_B, crabnet_kwargs,roost_kwargs,
random_state=seed)
num_results = [output[task][metric_reg] if ('regression' in task)
else output[task][metric_class] for task in tasks_list]
cols = results.columns.get_level_values('model')=='baseline'
results.loc[(prop,n),cols] = num_results
"""TRANSFER"""
if 'transfer' in models_list:
print(f'--- Transfer ---')
train_feat = utils.featurize(train, elem_prop=elem_prop)
output, _ = tasks.apply_all_tasks(train_feat, test_feat, key_A,
tasks_list, prop, key_B, crabnet_kwargs,roost_kwargs,
random_state=seed)
num_results = [output[task][metric_reg] if ('regression' in task)
else output[task][metric_class] for task in tasks_list]
cols = results.columns.get_level_values('model')=='transfer'
results.loc[(prop,n),cols] = num_results
"""CONCAT MODEL"""
if 'concat' in models_list:
print(f'--- concat ---')
# CONCATENATE
train_concat = concat(dfs_dict={key_A:train, key_B:data_B}, merging_opt=merging,
ascending=ascending_setting[prop])
# FEATURIZE TRAIN
train_feat = utils.featurize(train_concat, elem_prop=elem_prop)
# TASKS
output, _ = tasks.apply_all_tasks(train_feat, test_feat, key_A,
tasks_list, prop, key_B, crabnet_kwargs, roost_kwargs,
random_state=seed)
num_results = [output[task][metric_reg] if ('regression' in task)
else output[task][metric_class] for task in tasks_list]
cols = results.columns.get_level_values('model')=='concat'
results.loc[(prop,n),cols] = num_results
"""ELEM CONCAT MODEL"""
if 'elem_concat' in models_list:
print(f'--- elem concat ---')
# CONCATENATE
train_concat = elem_concat(dfs_dict={key_A:train, key_B:data_B}, merging_opt=merging,
ascending=ascending_setting[prop],
k=k_elemconcat, n=n_elemconcat, verbose=False,
random_state=seed)
# FEATURIZE TRAIN
train_feat = utils.featurize(train_concat, elem_prop=elem_prop)
# TASKS
output, _ = tasks.apply_all_tasks(train_feat, test_feat, key_A,
tasks_list, crabnet_kwargs,
random_state=seed)
num_results = [output[task][metric_reg] if ('regression' in task)
else output[task][metric_class] for task in tasks_list]
cols = results.columns.get_level_values('model')=='elem_concat'
results.loc[(prop,n),cols] = num_results
"""DISCOVER MODEL"""
if 'disco' in models_list:
print(f'--- discover ---')
# CONCATENATE
dam = DiscoAugment(dfs_dict={key_A:train, key_B:data_B},
self_augment_frac = None, # initial fraction for self_aumgent
random_state = seed)
aug_list = dam.apply_augmentation(crabnet_kwargs=crabnet_kwargs,
**discover_kwargs)
train_discoaug = aug_list[-1]
# FEATURIZE TRAIN
train_feat = utils.featurize(train_discoaug, elem_prop=elem_prop)
# TASKS
output, _ = tasks.apply_all_tasks(train_feat, test_feat, key_A,
tasks_list, crabnet_kwargs,
random_state=seed)
num_results = [output[task][metric_reg] if ('regression' in task)
else output[task][metric_class] for task in tasks_list]
cols = results.columns.get_level_values('model')=='disco'
results.loc[(prop,n),cols] = num_results
# saving results
with open(f'results/results_3_{prop}_transfer_mpds.pkl', 'wb') as handle:
pickle.dump(results, handle)
#average across repetitions
results_mean = results.groupby('prop').mean()
results_std = results.groupby('prop').std()
results_table = round(results_mean,3).astype(str) + " ± " + round(results_std,3).astype(str)
print('\n')
from tabulate import tabulate
h = [' '] + list(map('\n'.join, results.columns.tolist()))
content = tabulate(results_table, headers=h)
print(content)
# text_file=open("3_output_mae_new.csv","w")
# text_file.write(content)
# text_file.close()
if __name__ == '__main__': plot_all()