-
Notifications
You must be signed in to change notification settings - Fork 4
/
automl.py
executable file
·309 lines (231 loc) · 10.2 KB
/
automl.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
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
# %% [markdown]
# # Radiomics model building using Auto ML
# Wookjin Choi <wookjin.choi@jefferson.edu>
# %% [markdown]
# ## Import modules
# %%
import autosklearn
import autosklearn.classification
import sklearn.model_selection
import sklearn.datasets
import sklearn.metrics
from sklearn import datasets, cluster
from sklearn.tree import DecisionTreeClassifier
from sklearn.feature_selection import GenericUnivariateSelect
from sklearn.feature_selection import RFE
from sklearn.feature_selection import RFECV
from sklearn.feature_selection import mutual_info_classif
from sklearn.feature_selection import SelectFromModel
import pandas as pd
import numpy as np
# %% [markdown]
# ## 1. Data loading
# %%
clinical_lidc72 = pd.read_csv("metadata/LIDC_72.csv")
clinical_lidc = pd.read_csv("metadata/LIDC.csv")
clinical_lungx = pd.read_csv("metadata/LUNGx.csv")
#features_lidc = pd.read_csv("output/feature-list_nodule-lidc_pylidc.csv")
features_lidc = pd.read_csv("output/feature-list_nodule-lidc.csv")
features_lungx = pd.read_csv("output/feature-list_nodule-lungx.csv")
lidc_tags = features_lidc.Tags.str.split("-").apply(pd.Series)
features_lidc['NID'] = lidc_tags[0].astype(int)
features_lidc['Tags'] = lidc_tags[1]
lungx_tags = features_lungx.Tags.str.split("-").apply(pd.Series)
lungx_tags.iloc[lungx_tags[2].isna(), 1] = "gc"
#lungx_tags.iloc[lungx_tags[2].isna(), 2] = "1mm"
features_lungx['NID'] = lungx_tags[0].astype(int)
features_lungx['Tags'] = lungx_tags[1]
features_lidc = features_lidc.loc[(features_lidc.Tags=='seg')]
features_lungx = features_lungx.loc[(features_lungx.Tags=='seg')]
#features_lidc = features_lidc.drop(columns=["WeightedElongation","WeightedFlatness","2DWeightedElongation","2DWeightedFlatness"])
#features_lungx = features_lungx.drop(columns=["WeightedElongation","WeightedFlatness","2DWeightedElongation","2DWeightedFlatness"])
features_lidc72 = features_lidc.loc[features_lidc.PID.isin(clinical_lidc72.PID) & (features_lidc.NID==1)]
features_lidc72RM = features_lidc.loc[features_lidc.PID.isin(clinical_lidc72.PID)]
clinical_lidc = clinical_lidc.loc[clinical_lidc.set_index(['PID','NID']).index.isin(features_lidc.set_index(['PID','NID']).index)]
features_lidc = features_lidc.loc[features_lidc.set_index(['PID','NID']).index.isin(clinical_lidc.set_index(['PID','NID']).index)]
features_lidc_train = features_lidc.loc[~features_lidc.PID.isin(clinical_lidc72.PID)]
clinical_lidc_train = clinical_lidc.loc[~clinical_lidc.PID.isin(clinical_lidc72.PID)]
clinical_lidc72RM = clinical_lidc.loc[clinical_lidc.PID.isin(clinical_lidc72.PID)]
# %%
spiculation_lidc = pd.read_csv("output/nodule-lidc/spiculation/features.csv")
spiculation_lungx = pd.read_csv("output/nodule-lungx/spiculation/features.csv")
spiculation_lidc72RM = spiculation_lidc.loc[spiculation_lidc.PID.isin(clinical_lidc72.PID)]
spiculation_lidc72 = spiculation_lidc.loc[spiculation_lidc.PID.isin(clinical_lidc72.PID) & (spiculation_lidc.NID==1)]
spiculation_lidc = spiculation_lidc.loc[spiculation_lidc.set_index(['PID','NID']).index.isin(clinical_lidc.set_index(['PID','NID']).index)]
spiculation_lidc_train = spiculation_lidc.loc[~spiculation_lidc.PID.isin(clinical_lidc72.PID)]
# %% [markdown]
# ### 1.1 Data preprocessing
# %%
#X = features_lidc_train.iloc[:, 4:-1].update(spiculation_lungx.iloc[:,3:])
X = pd.concat([features_lidc_train.reset_index().iloc[:, 5:-1], spiculation_lidc_train.reset_index().iloc[:,4:]], axis=1) # with spiculation features
X = X.loc[:,X.isna().sum(axis=0)==0]
y = clinical_lidc_train.Malignancy > 3
#X_72RM = features_lidc72RM.iloc[:, 4:-1]
X_72RM = pd.concat([features_lidc72RM.reset_index().iloc[:, 5:-1], spiculation_lidc72RM.reset_index().iloc[:,4:]], axis=1) # with spiculation features
X_72RM = X_72RM.loc[:,X_72RM.isna().sum(axis=0)==0]
y_72RM = clinical_lidc72RM.Malignancy > 3
#X_72 = features_lidc72.iloc[:, 4:-1]
X_72 = pd.concat([features_lidc72.reset_index().iloc[:, 5:-1], spiculation_lidc72.reset_index().iloc[:,4:]], axis=1) # with spiculation features
X_72 = X_72.loc[:,X_72.isna().sum(axis=0)==0]
y_72 = clinical_lidc72.PMalignancy == 2
#X_lungx = features_lungx.iloc[:, 4:-1]
X_lungx = pd.concat([features_lungx.reset_index().iloc[:, 5:-1], spiculation_lungx.reset_index().iloc[:,4:]], axis=1) # with spiculation features
X_lungx = X_lungx.loc[:,X_lungx.isna().sum(axis=0)==0]
y_lungx = clinical_lungx.malignancy > 0
# %%
columns = np.intersect1d(np.intersect1d(np.intersect1d(X.columns, X_72RM.columns), X_72.columns), X_lungx.columns)
X = X[columns]
X_72RM = X_72RM[columns]
X_72 = X_72[columns]
X_lungx = X_lungx[columns]
# %%
X_lungx_cal = X_lungx.iloc[0:10]
y_lungx_cal = y_lungx.iloc[0:10]
X_lungx_test = X_lungx.iloc[10:]
y_lungx_test = y_lungx.iloc[10:]
# %% [markdown]
# ### 1.2 Data splitting
# %%
X_train, X_test, y_train, y_test = \
sklearn.model_selection.train_test_split(X, y, test_size=0.2, random_state=42)
# %% [markdown]
# ## 2. Feature Selection
# %% [markdown]
# ### 2.1 Redundant Feature Elimination
# %%
agglo = cluster.FeatureAgglomeration(affinity='precomputed', linkage='complete', n_clusters=None, distance_threshold=0.3)
agglo.fit(1-abs(X_train.corr()))
cv_f = X_train.std()/X_train.mean().abs()
feature_groups = [np.array(X_train.columns[agglo.labels_==i]) for i in np.unique(agglo.labels_)]
feature_reps = [cv_f[f].sort_values().index[-1] for f in feature_groups]
feature_reps
# %%
#X_reduced_tr = pd.DataFrame(agglo.transform(X_train))
#X_reduced_ts = pd.DataFrame(agglo.transform(X_test))
#X_reduced_tr.columns = feature_reps
#X_reduced_ts.columns = feature_reps
X_reduced_tr = X_train[feature_reps]
X_reduced_ts = X_test[feature_reps]
print(X_reduced_tr.shape, X_reduced_ts.shape)
feature_groups[0].shape
# %%
X_train = X_reduced_tr
X_test = X_reduced_ts
# %% [markdown]
# ### 2.2 Feature Selection
# %%
trans = GenericUnivariateSelect(score_func=lambda X, y: X.mean(axis=0), mode='percentile', param=50)
X_trans = trans.fit_transform(X_train, y_train)
mutual_information = mutual_info_classif(X_train, y_train)
trans = GenericUnivariateSelect(score_func=mutual_info_classif, mode='k_best', param=30)
X_trans = trans.fit_transform(X_train, y_train)
columns_retained_Select = X_train.columns[trans.get_support()].values
columns_retained_Select
# %%
clf = DecisionTreeClassifier()
clf.fit(X, y)
trans = SelectFromModel(clf, threshold='median', max_features=30)
X_trans = trans.fit_transform(X, y)
columns_retained_FromMode = X.columns[trans.get_support()].values
columns_retained_FromMode
# %%
clf = DecisionTreeClassifier()
trans = RFE(clf, n_features_to_select=30)
X_trans = trans.fit_transform(X_train, y_train)
columns_retained_RFE = X_train.columns[trans.get_support()].values
columns_retained_RFE
# %%
clf = DecisionTreeClassifier()
trans = RFECV(clf)
X_trans = trans.fit_transform(X_train, y_train)
columns_retained_RFECV = X_train.columns[trans.get_support()].values
columns_retained_RFECV
# %%
columns_retained = columns_retained_RFE
# %% [markdown]
# ## 3. AutoML
# %% [markdown]
# ### 3.1 TPOT
# %%
from tpot import TPOTClassifier
# X_train, X_test, y_train, y_test = \
# sklearn.model_selection.train_test_split(X, y, random_state=42)
pipeline_optimizer = TPOTClassifier(generations=5, population_size=300, cv=5,
random_state=42, verbosity=2)
# pipeline_optimizer.fit(X_train, y_train)
# print(pipeline_optimizer.score(X_test, y_test))
pipeline_optimizer.fit(X_train[columns_retained], y_train)
#pipeline_optimizer.export('tpot_exported_pipeline.py')
# %% [markdown]
# #### 3.2.1 TPOT validation
# %%
print("Test AUC:", pipeline_optimizer.score(X_test[columns_retained], y_test))
print("72RM AUC:", pipeline_optimizer.score(X_72RM[columns_retained], y_72RM))
print("72 AUC:", pipeline_optimizer.score(X_72[columns_retained], y_72))
print("LUNGx AUC:", pipeline_optimizer.score(X_lungx_test[columns_retained], y_lungx_test))
# %% [markdown]
# ### 3.2 AutoSklearn
# %%
automl = autosklearn.classification.AutoSklearnClassifier(
time_left_for_this_task = 8000,
per_run_time_limit = 800,
seed=42,
memory_limit=None,
metric=autosklearn.metrics.roc_auc,
resampling_strategy='cv',
resampling_strategy_arguments=dict(folds=5),
)
automl.fit(X_train[columns_retained], y_train)
print(np.max(automl.cv_results_['mean_test_score']))
# %% [markdown]
# #### 3.2.1 AutoSklearn validation
# %%
#predict
y_hat = automl.predict(X_test[columns_retained])
y_pred = automl.predict_proba(X_test[columns_retained])
y_hat_72RM = automl.predict(X_72RM[columns_retained])
y_pred_72RM = automl.predict_proba(X_72RM[columns_retained])
y_hat_72 = automl.predict(X_72[columns_retained])
y_pred_72 = automl.predict_proba(X_72[columns_retained])
y_hat_lungx = automl.predict(X_lungx_test[columns_retained])
y_pred_lungx = automl.predict_proba(X_lungx_test[columns_retained])
# show scores
print("Test AUC: ", sklearn.metrics.roc_auc_score(y_test, y_pred[:,1]))
print("72RM AUC: ", sklearn.metrics.roc_auc_score(y_72RM, y_pred_72RM[:,1]))
print("72 AUC: ", sklearn.metrics.roc_auc_score(y_72, y_pred_72[:,1]))
print("LUNGx AUC: ", sklearn.metrics.roc_auc_score(y_lungx_test, y_pred_lungx[:,1]))
print("\nTest Accuracy: ", sklearn.metrics.accuracy_score(y_test, y_hat))
print("72RM Accuracy: ", sklearn.metrics.accuracy_score(y_72RM, y_hat_72RM))
print("72 Accuracy: ", sklearn.metrics.accuracy_score(y_72, y_hat_72))
print("LUNGx Accuracy: ", sklearn.metrics.accuracy_score(y_lungx_test, y_hat_lungx))
# %% [markdown]
# #### 3.2.2 AutoSklearn models
# %%
# show all models
show_models_str=automl.show_models()
sprint_statistics_str = automl.sprint_statistics()
leaderboard = automl.leaderboard()
print(sprint_statistics_str)
print(leaderboard)
# %%
print(show_models_str[leaderboard.index[0]])
# %%
import matplotlib.pyplot as plt
poT = automl.performance_over_time_
poT.plot(
x='Timestamp',
kind='line',
legend=True,
title='Auto-sklearn AUC over time',
grid=True,
)
plt.show()
# %%
def get_metric_result(cv_results):
results = pd.DataFrame.from_dict(cv_results)
results = results[results['status'] == "Success"]
cols = ['rank_test_scores', 'param_classifier:__choice__', 'mean_test_score']
cols.extend([key for key in cv_results.keys() if key.startswith('metric_')])
return results[cols]
print("Metric results")
print(get_metric_result(automl.cv_results_).to_string(index=False))