-
Notifications
You must be signed in to change notification settings - Fork 74
/
ml_lu.py
248 lines (214 loc) · 10.1 KB
/
ml_lu.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
"""
Implementation of prior depression detection algorithm:
Jin Lu, Jinbo Bi, Chao Shang, Chaoqun Yue, Reynaldo Morillo, Shweta Ware,
Jayesh Kamath, Athanasios Bamis, Alexander Russell, and Bing Wang. 2018.
Joint Modeling of Heterogeneous Sensing Data for Depression Assessment via Multi-task Learning.
Proceedings of the ACM on Interactive, Mobile,Wearable and Ubiquitous Technologies 2, 1 (2018) 1–21
https://doi.org/10.1145/3191753 ISBN: 9781450351980.
"""
import os, sys
sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
from utils.common_settings import *
from algorithm.ml_basic import DepressionDetectionAlgorithm_ML_basic
from data_loader.data_loader_ml import DataRepo
from algorithm.base import DepressionDetectionClassifierBase
from utils.cv_split import judge_corner_cvsplit
class DepressionDetectionClassifier_ML_lu(DepressionDetectionClassifierBase):
"""Classifier for Lu et al. work. Fits two models based on device type with a MTL framework"""
def __init__(self, gamma1, gamma2, selected_features, p = 1, k = 2):
self.device_type_set = set([0,1]) # ios - 1, android - 0
self.selected_features = selected_features
self.num_tasks = 2
self.tolerance = 0.01
self.max_iter = 10
self.gamma1 = gamma1
self.gamma2 = gamma2
self.p = p
self.k = k
assert self.p in [1,2]
assert self.k in [1,2]
self.c_history = []
self.alpha_history = []
self.beta_history = []
self.no_data_flag_ios = False
self.no_data_flag_android = False
def fit(self, X, y=None):
assert set(self.selected_features).issubset(set(X.columns))
assert X.columns[-1] == "device_type"
set_random_seed(42)
X_np = np.array(X)
y_np = np.array(y)
devices = X_np[:,-1]
assert set(devices).issubset(self.device_type_set)
self.feature_num = X_np.shape[-1] - 1
devices_ios_index = np.where(devices == 1)[0]
devices_android_index = np.where(devices != 1)[0]
# Task 1
X_ios = X_np[devices_ios_index,:-1]
y_ios = y_np[devices_ios_index]
# Task 2
X_android = X_np[devices_android_index,:-1]
y_android = y_np[devices_android_index]
if (len(devices_ios_index) == 0):
self.no_data_flag_ios = True
else:
self.no_data_flag_ios = False
if (len(devices_android_index) == 0):
self.no_data_flag_android = True
else:
self.no_data_flag_android = False
X_tasks = []
y_tasks = []
if (not self.no_data_flag_ios):
X_tasks.append(X_ios)
y_tasks.append(y_ios)
if (not self.no_data_flag_android):
X_tasks.append(X_android)
y_tasks.append(y_android)
self.c_diag = np.ones(self.feature_num)
self.c = np.diag(self.c_diag)
self.c_history.append(deepcopy(self.c_diag))
self.blockwise_coordinate_descent(X_tasks, y_tasks)
return self.find_best_threshold(X_tasks, y_tasks)
def blockwise_coordinate_descent(self, X_tasks, y_tasks):
while True:
beta_tasks = []
alpha_tasks = []
for task in range(len(X_tasks)):
X_t = X_tasks[task]
y_t = y_tasks[task]
X_t_hat = np.matmul(X_t, self.c)
clf = linear_model.LogisticRegression(penalty = "l1" if self.p == 1 else "l2",
C = self.gamma1, solver = "saga", random_state = 42,
max_iter=30000, tol = 0.001)
clf.fit(X_t_hat, y_t)
beta_t = deepcopy(clf.coef_).T
alpha_t = np.matmul(self.c, beta_t)
beta_tasks.append(beta_t)
alpha_tasks.append(alpha_t)
alpha_tasks = np.array(alpha_tasks)
self.beta = np.matrix(np.concatenate(beta_tasks, axis = 1))
self.alpha = np.matrix(np.concatenate(alpha_tasks, axis = 1))
self.beta_history.append(self.beta)
self.alpha_history.append(self.alpha)
c_diag_new = np.empty(len(self.c))
s = (alpha_tasks ** self.p).sum(axis = 0)
c_diag_new = (self.gamma1 * np.abs(s) / self.gamma2) ** (1.0 / (self.p + self.k)) * np.sign(s)
self.c_diag = c_diag_new[:,0]
self.c = np.diag(self.c_diag)
self.c_history.append(deepcopy(self.c_diag))
if (len(self.alpha_history) > 2):
if (len(self.alpha_history) > self.max_iter):
# print("Exceed max iter!", self.gamma1, self.gamma2)
break
delta = np.linalg.norm(self.alpha - self.alpha_history[-2], ord = 1)
if (delta < self.tolerance):
# print("iter num", len(self.alpha_history))
break
def find_best_threshold(self, X_tasks, y_tasks):
y_prob_train = []
y_test = []
best_th_tasks = []
for task in range(len(X_tasks)):
X_t = X_tasks[task]
y_t = y_tasks[task]
alpha = self.alpha_history[-1][:,task]
y_prob = 1 / (1 + np.exp(-np.asarray(np.matmul(X_t, alpha))[:,0]))
y_prob_train.append(y_prob)
y_test.append(y_t)
roc = sklearn.metrics.roc_curve(y_true = y_t, y_score = y_prob)
best_th = roc[2][np.argmax((1 - roc[0] + roc[1]) / 2)] # select based on bal acc
best_th_tasks.append(best_th)
y_prob_train = np.concatenate(y_prob_train)
y_test = np.concatenate(y_test)
roc = sklearn.metrics.roc_curve(y_true = y_test, y_score = y_prob_train)
best_th = roc[2][np.argmax((1 - roc[0] + roc[1]) / 2)] # select based on bal acc
self.best_th_single = best_th
self.best_th_tasks = deepcopy(best_th_tasks)
def predict_complete(self, X, y=None):
X_np = np.array(X)
y_pred = np.empty(len(X_np))
y_pred_prob = np.empty((len(X_np),2))
devices = X_np[:,-1]
devices_ios_index = np.where(devices == 1)[0]
devices_android_index = np.where(devices != 1)[0]
X_np = np.array(X)
y_np = np.array(y)
devices = X_np[:,-1]
devices_ios_index = np.where(devices == 1)[0]
devices_android_index = np.where(devices != 1)[0]
X_ios = X_np[devices_ios_index,:-1]
X_android = X_np[devices_android_index,:-1]
X_tasks = []
idx_tasks = []
if (len(devices_ios_index) > 0):
X_tasks.append(X_ios)
idx_tasks.append(devices_ios_index)
if (len(devices_android_index) > 0):
X_tasks.append(X_android)
idx_tasks.append(devices_android_index)
for task, (X_t, idx_t) in enumerate(zip(X_tasks, idx_tasks)):
alpha = self.alpha_history[-1][:,task]
y_prob = 1 / (1 + np.exp(-np.asarray(np.matmul(X_t, alpha))[:,0]))
best_th = self.best_th_tasks[task]
y_pred[idx_t] = y_prob > best_th
y_prob_tmp = 1 / (1 + np.exp(-np.asarray(np.matmul(X_t, alpha))[:,0]))
y_prob_list = np.concatenate([[1 - y_prob_tmp], [y_prob_tmp]]).T
y_pred_prob[idx_t] = y_prob_list
return y_pred, y_pred_prob
def predict(self, X, y=None):
y_pred, y_pred_prob = self.predict_complete(X, y)
return y_pred
def predict_proba(self, X, y=None):
y_pred, y_pred_prob = self.predict_complete(X, y)
return y_pred_prob
class DepressionDetectionAlgorithm_ML_lu(DepressionDetectionAlgorithm_ML_basic):
"""Algirithm for Lu et al. work, extending the basic traditional ml algorithm """
def __init__(self, config_dict = None, config_name = "ml_lu"):
super().__init__(config_dict, config_name)
def prep_model(self, data_train: DataRepo, criteria: str = "f1") -> sklearn.base.ClassifierMixin:
super().prep_model()
set_random_seed(42)
@ray.remote
def train_small_cv(data_repo: DataRepo, model_parameters: dict):
warnings.filterwarnings("ignore")
X = data_repo.X
y = data_repo.y
pids = data_repo.pids
model_parameters_full = deepcopy(model_parameters)
model_parameters_full.update({"selected_features": X.columns})
clf = DepressionDetectionClassifier_ML_lu(**model_parameters_full)
cv = StratifiedGroupKFold(n_splits=5, shuffle=True, random_state=42)
repeat_time = 0
while True:
repeat_time += 1
cv = StratifiedGroupKFold(n_splits=5,shuffle=True, random_state=42+repeat_time*1000)
if (judge_corner_cvsplit(cv, data_repo)):
continue
else:
break
r = cross_validate(clf, X=X, y=y, groups=pids, cv = cv,
scoring = utils_ml.results_report_sklearn,
return_train_score=False)
r = {k:np.mean(v) for k,v in r.items()}
r.update({"parameters":model_parameters})
return r
gamma1_list = [10**i for i in range(-2,3)]
gamma2_list = [10**i for i in range(-2,3)]
p_list = [1] # [1,2]
k_list = [1] # [2,1]
parameters_list = []
for gamma1, gamma2, p, k in itertools.product(gamma1_list, gamma2_list, p_list, k_list):
parameters_tmp = {"gamma1": gamma1, "gamma2": gamma2, "p":p, "k":k}
parameters_list.append(parameters_tmp)
data_train_id = ray.put(data_train)
results_list = ray.get([train_small_cv.remote(data_train_id,i) for i in parameters_list])
results_list = pd.DataFrame(results_list)
best_row = results_list.iloc[results_list[f"test_{criteria}"].argmax()]
best_params = best_row['parameters']
if (self.verbose > 0):
print(best_row)
print(best_params)
best_params_full = deepcopy(best_params)
best_params_full.update({"selected_features": data_train.X.columns})
return DepressionDetectionClassifier_ML_lu(**best_params_full)