-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_simple_Tchange.py
334 lines (278 loc) · 15.9 KB
/
run_simple_Tchange.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
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
from optparse import OptionParser
op = OptionParser()
op.add_option("-M", "--Ngroups", type=int, default=3, help="number of groups in propose formulation")
op.add_option("-p", "--path", type="string", default='data/', help="path for data (path/synthetic..)")
(opts, args) = op.parse_args()
path = opts.path
M_seted = opts.Ngroups
state_sce = None
#GLOBAL Variables
BATCH_SIZE = 64 #128
EPOCHS_BASE = 50
OPT = 'adam' #optimizer for neural network
TOL = 3e-2 #tolerance for relative variation of parameters
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import keras, time, sys, os, gc
from sklearn.metrics import confusion_matrix
from keras.models import clone_model
DTYPE_OP = 'float32'
keras.backend.set_floatx(DTYPE_OP)
if DTYPE_OP == 'float64':
keras.backend.set_epsilon(np.finfo(np.float64).eps)
elif DTYPE_OP == 'float32':
keras.backend.set_epsilon(np.finfo(np.float32).eps)
### Load Data
X_train = np.loadtxt(path+"/synthetic/simple/datasim_X_train.csv",delimiter=',')
Z_train = np.loadtxt(path+"/synthetic/simple/datasim_Z_train.csv",dtype='int') #groudn truth
X_test = np.loadtxt(path+"/synthetic/simple/datasim_X_test.csv",delimiter=',')
Z_test = np.loadtxt(path+"/synthetic/simple/datasim_Z_test.csv",dtype='int') #groudn truth
print("Input shape:",X_train.shape)
from sklearn.preprocessing import StandardScaler
std= StandardScaler(with_mean=True) #matrices sparse with_mean=False
std.fit(X_train)
Xstd_train = std.transform(X_train)
Xstd_test = std.transform(X_test)
from code.learning_models import LogisticRegression_Sklearn,LogisticRegression_Keras,MLP_Keras
from code.learning_models import default_CNN,default_RNN,default_RNNw_emb,CNN_simple, RNN_simple #deep learning
from code.evaluation import Evaluation_metrics
from code.representation import *
from code.utils import *
from code.baseline import LabelInference, RaykarMC
from code.MixtureofGroups import GroupMixtureOpt, project_and_cluster,clusterize_annotators
### Delta convergence criteria
from code.utils import EarlyStopRelative
ourCallback = EarlyStopRelative(monitor='loss',patience=1,min_delta=TOL)
start_time_exec = time.time()
#upper bound model
Z_train_onehot = keras.utils.to_categorical(Z_train)
model_UB = MLP_Keras(Xstd_train.shape[1:],Z_train_onehot.shape[1],16,1,BN=False,drop=0.2) #what about bn true?
model_UB.compile(loss='categorical_crossentropy',optimizer=OPT)
hist = model_UB.fit(Xstd_train,Z_train_onehot,epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])
print("Trained Ideal Model , Epochs to converge =",len(hist.epoch))
evaluate = Evaluation_metrics(model_UB,'keras',Xstd_train.shape[0],plot=False)
Z_train_pred = model_UB.predict_classes(Xstd_train)
results1 = evaluate.calculate_metrics(Z=Z_train,Z_pred=Z_train_pred)
Z_test_pred = model_UB.predict_classes(Xstd_test)
results2 = evaluate.calculate_metrics(Z=Z_test,Z_pred=Z_test_pred)
results1[0].to_csv("synthetic_UpperBound_train.csv",index=False)
results2[0].to_csv("synthetic_UpperBound_test.csv",index=False)
del evaluate,Z_train_pred,Z_test_pred,results1,results2
gc.collect()
keras.backend.clear_session()
def get_mean_dataframes(df_values):
if df_values[0].iloc[:,0].dtype == object:
RT = pd.DataFrame(data=None,columns = df_values[0].columns[1:], index= df_values[0].index)
else:
RT = pd.DataFrame(data=None,columns = df_values[0].columns, index= df_values[0].index)
data = []
for df_value in df_values:
if df_value.iloc[:,0].dtype == object:
data.append( df_value.iloc[:,1:].values )
else:
data.append(df_value.values)
RT[:] = np.mean(data,axis=0)
if df_values[0].iloc[:,0].dtype == object:
RT.insert(0, "", df_values[0].iloc[:,0].values )
return RT
from code.generate_data import SinteticData
#ANNOTATOR DENSITY CHOOSE
to_check = [100,500,1500,3500,6000,10000]
T_data = 5
results_softmv_train = []
results_softmv_test = []
results_hardmv_train = []
results_hardmv_test = []
results_ds_train = []
results_ds_test = []
results_raykar_train = []
results_raykar_trainA = []
results_raykar_test = []
results_ours_global_train = []
results_ours_global_trainA = []
results_ours_global_test = []
results_ours_global_testA = []
for Tmax in to_check:
aux_softmv_train = []
aux_softmv_test = []
aux_hardmv_train = []
aux_hardmv_test = []
aux_ds_train = []
aux_ds_test = []
aux_raykar_train = []
aux_raykar_trainA = []
aux_raykar_test = []
aux_ours_global_train = []
aux_ours_global_trainA = []
aux_ours_global_test = []
aux_ours_global_testA = []
GenerateData = SinteticData(state=state_sce) #por la semilla quedan similares..
#CONFUSION MATRIX CHOOSE
GenerateData.set_probas(asfile=True,file_matrix=path+'/synthetic/simple/matrix_datasim_normal.csv',file_groups =path+'/synthetic/simple/groups_datasim_normal.csv')
real_conf_matrix = GenerateData.conf_matrix.copy()
print("New Synthetic data is being generated...",flush=True,end='')
y_obs, groups_annot = GenerateData.sintetic_annotate_data(Z_train,Tmax,T_data,deterministic=False,hard=True)
print("Done! ")
if len(groups_annot.shape) ==1 or groups_annot.shape[1] == 1:
groups_annot = keras.utils.to_categorical(groups_annot) #only if it is hard clustering
confe_matrix_R = np.tensordot(groups_annot,real_conf_matrix, axes=[[1],[0]])
T_weights = np.sum(y_obs != -1,axis=0)
print("Mean annotations by t= ",T_weights.mean())
N,T = y_obs.shape
K = np.max(y_obs)+1 # asumiendo que estan ordenadas
print("Shape (data,annotators): ",(N,T))
print("Classes: ",K)
############### MV/DS and calculate representations##############################
label_I = LabelInference(y_obs,TOL,type_inf = 'all') #Infer Labels
mv_probas, mv_conf_probas = label_I.mv_labels('probas')
mv_onehot, mv_conf_onehot = label_I.mv_labels('onehot')
confe_matrix_G = get_Global_confusionM(Z_train,label_I.y_obs_repeat)
#Deterministic
ds_labels, ds_conf = label_I.DS_labels()
print("ACC MV on train:",np.mean(mv_onehot.argmax(axis=1)==Z_train))
print("ACC D&S on train:",np.mean(ds_labels.argmax(axis=1)==Z_train))
#get representation needed for Raykar
start_time = time.time()
y_obs_categorical = set_representation(y_obs,'onehot')
print("shape:",y_obs_categorical.shape)
print("Representation for Raykar in %f mins"%((time.time()-start_time)/60.) )
#get our global representation
r_obs = label_I.y_obs_repeat.copy() #set_representation(y_obs_categorical,"repeat")
print("vector of repeats:\n",r_obs)
print("shape:",r_obs.shape)
for _ in range(30): #repetitions
############# EXECUTE ALGORITHMS #############################
model_mvsoft = clone_model(model_UB)
model_mvsoft.compile(loss='categorical_crossentropy',optimizer=OPT)
hist = model_mvsoft.fit(Xstd_train, mv_probas, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])
print("Trained model over soft-MV, Epochs to converge =",len(hist.epoch))
Z_train_pred_mvsoft = model_mvsoft.predict_classes(Xstd_train)
Z_test_pred_mvsoft = model_mvsoft.predict_classes(Xstd_test)
keras.backend.clear_session()
model_mvhard = clone_model(model_UB)
model_mvhard.compile(loss='categorical_crossentropy',optimizer=OPT)
hist=model_mvhard.fit(Xstd_train, mv_onehot, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])
print("Trained model over hard-MV, Epochs to converge =",len(hist.epoch))
Z_train_pred_mvhard = model_mvhard.predict_classes(Xstd_train)
Z_test_pred_mvhard = model_mvhard.predict_classes(Xstd_test)
keras.backend.clear_session()
model_ds = clone_model(model_UB)
model_ds.compile(loss='categorical_crossentropy',optimizer=OPT)
hist=model_ds.fit(Xstd_train, ds_labels, epochs=EPOCHS_BASE,batch_size=BATCH_SIZE,verbose=0,callbacks=[ourCallback])
print("Trained model over D&S, Epochs to converge =",len(hist.epoch))
Z_train_pred_ds = model_ds.predict_classes(Xstd_train)
Z_test_pred_ds = model_ds.predict_classes(Xstd_test)
keras.backend.clear_session()
raykarMC = RaykarMC(Xstd_train.shape[1:],y_obs_categorical.shape[-1],T,epochs=1,optimizer=OPT,DTYPE_OP=DTYPE_OP)
raykarMC.define_model('mlp',16,1,BatchN=False,drop=0.2)
logL_hists,i_r = raykarMC.multiples_run(20,Xstd_train,y_obs_categorical,batch_size=BATCH_SIZE,max_iter=EPOCHS_BASE,tolerance=TOL)
Z_train_p_Ray = raykarMC.base_model.predict(Xstd_train)
Z_test_pred_Ray = raykarMC.base_model.predict_classes(Xstd_test)
keras.backend.clear_session()
gMixture_Global = GroupMixtureOpt(Xstd_train.shape[1:],Kl=r_obs.shape[1],M=M_seted,epochs=1,pre_init=0,optimizer=OPT,dtype_op=DTYPE_OP)
gMixture_Global.define_model("mlp",16,1,BatchN=False,drop=0.2)
gMixture_Global.lambda_random = False #with lambda random --necessary
logL_hists,i = gMixture_Global.multiples_run(20,Xstd_train,r_obs,batch_size=BATCH_SIZE,max_iter=EPOCHS_BASE,tolerance=TOL
,cluster=True)
Z_train_p_OG = gMixture_Global.base_model.predict(Xstd_train)
Z_test_p_OG = gMixture_Global.base_model.predict(Xstd_test)
keras.backend.clear_session()
################## MEASURE PERFORMANCE ##################################
evaluate = Evaluation_metrics(model_mvsoft,'keras',Xstd_train.shape[0],plot=False)
evaluate.set_T_weights(T_weights)
prob_Yzt = np.tile( mv_conf_probas, (T,1,1) )
results1 = evaluate.calculate_metrics(Z=Z_train,Z_pred=Z_train_pred_mvsoft,conf_pred=prob_Yzt,conf_true=confe_matrix_R,
conf_true_G =confe_matrix_G, conf_pred_G = mv_conf_probas)
results2 = evaluate.calculate_metrics(Z=Z_test,Z_pred=Z_test_pred_mvsoft)
aux_softmv_train += results1
aux_softmv_test += results2
evaluate = Evaluation_metrics(model_mvhard,'keras',Xstd_train.shape[0],plot=False)
evaluate.set_T_weights(T_weights)
prob_Yzt = np.tile( mv_conf_onehot, (T,1,1) )
results1 = evaluate.calculate_metrics(Z=Z_train,Z_pred=Z_train_pred_mvhard,conf_pred=prob_Yzt,conf_true=confe_matrix_R,
conf_true_G =confe_matrix_G, conf_pred_G = mv_conf_onehot)
results2 = evaluate.calculate_metrics(Z=Z_test,Z_pred=Z_test_pred_mvhard)
aux_hardmv_train += results1
aux_hardmv_test += results2
evaluate = Evaluation_metrics(model_ds,'keras',Xstd_train.shape[0],plot=False)
evaluate.set_T_weights(T_weights)
results1 = evaluate.calculate_metrics(Z=Z_train,Z_pred=Z_train_pred_ds,conf_pred=ds_conf,conf_true=confe_matrix_R,
conf_true_G =confe_matrix_G, conf_pred_G = ds_conf.mean(axis=0))
results2 = evaluate.calculate_metrics(Z=Z_test,Z_pred=Z_test_pred_ds)
aux_ds_train += results1
aux_ds_test += results2
evaluate = Evaluation_metrics(raykarMC,'raykar',plot=False)
prob_Yzt = raykarMC.get_confusionM()
prob_Yxt = raykarMC.get_predictions_annot(Xstd_train,data=Z_train_p_Ray)
Z_train_pred_Ray = Z_train_p_Ray.argmax(axis=-1)
results1 = evaluate.calculate_metrics(Z=Z_train,Z_pred=Z_train_pred_Ray,conf_pred=prob_Yzt,conf_true=confe_matrix_R,
y_o=y_obs,yo_pred=prob_Yxt,conf_true_G =confe_matrix_G, conf_pred_G = prob_Yzt.mean(axis=0))
results1_aux = evaluate.calculate_metrics(y_o=y_obs,yo_pred=prob_Yxt)
results2 = evaluate.calculate_metrics(Z=Z_test,Z_pred=Z_test_pred_Ray)
aux_raykar_train += results1
aux_raykar_trainA += results1_aux
aux_raykar_test += results2
evaluate = Evaluation_metrics(gMixture_Global,'our1',plot=False)
aux = gMixture_Global.calculate_extra_components(Xstd_train,y_obs,T=T,calculate_pred_annotator=True,p_z=Z_train_p_OG)
predictions_m,prob_Gt,prob_Yzt,prob_Yxt = aux #to evaluate...
prob_Yz = gMixture_Global.calculate_Yz()
Z_train_pred_OG = Z_train_p_OG.argmax(axis=-1)
results1 = evaluate.calculate_metrics(Z=Z_train,Z_pred=Z_train_pred_OG,conf_pred=prob_Yzt,conf_true=confe_matrix_R,
y_o=y_obs,yo_pred=prob_Yxt,
conf_true_G =confe_matrix_G, conf_pred_G = prob_Yz)
results1_aux = evaluate.calculate_metrics(y_o=y_obs,yo_pred=prob_Yxt)
c_M = gMixture_Global.get_confusionM()
y_o_groups = gMixture_Global.get_predictions_groups(Xstd_test,data=Z_test_p_OG).argmax(axis=-1) #obtain p(y^o|x,g=m) and then argmax
Z_test_pred_OG = Z_test_p_OG.argmax(axis=-1)
results2 = evaluate.calculate_metrics(Z=Z_test,Z_pred=Z_test_pred_OG,conf_pred=c_M, y_o_groups=y_o_groups)
aux_ours_global_train += results1
aux_ours_global_trainA += results1_aux
aux_ours_global_testA.append(results2[0])
aux_ours_global_test.append(results2[1])
print("All Performance Measured")
del model_mvsoft,model_mvhard,model_ds,raykarMC,gMixture_Global,evaluate
gc.collect()
#plot measures
results_softmv_train.append(get_mean_dataframes(aux_softmv_train))
results_softmv_test.append(get_mean_dataframes(aux_softmv_test))
results_hardmv_train.append(get_mean_dataframes(aux_hardmv_train))
results_hardmv_test.append(get_mean_dataframes(aux_hardmv_test))
results_ds_train.append(get_mean_dataframes(aux_ds_train))
results_ds_test.append(get_mean_dataframes(aux_ds_test))
results_raykar_train.append(get_mean_dataframes(aux_raykar_train))
results_raykar_trainA.append(get_mean_dataframes(aux_raykar_trainA))
results_raykar_test.append(get_mean_dataframes(aux_raykar_test))
results_ours_global_train.append(get_mean_dataframes(aux_ours_global_train))
results_ours_global_trainA.append(get_mean_dataframes(aux_ours_global_trainA))
results_ours_global_test.append(get_mean_dataframes(aux_ours_global_test))
results_ours_global_testA.append(get_mean_dataframes(aux_ours_global_testA))
gc.collect()
import pickle
with open('synthetic_softMV_train.pickle', 'wb') as handle:
pickle.dump(results_softmv_train, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_softMV_test.pickle', 'wb') as handle:
pickle.dump(results_softmv_test, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_hardMV_train.pickle', 'wb') as handle:
pickle.dump(results_hardmv_train, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_hardMV_test.pickle', 'wb') as handle:
pickle.dump(results_hardmv_test, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_DS_train.pickle', 'wb') as handle:
pickle.dump(results_ds_train, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_DS_test.pickle', 'wb') as handle:
pickle.dump(results_ds_test, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_Raykar_train.pickle', 'wb') as handle:
pickle.dump(results_raykar_train, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_Raykar_trainAnn.pickle', 'wb') as handle:
pickle.dump(results_raykar_trainA, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_Raykar_test.pickle', 'wb') as handle:
pickle.dump(results_raykar_test, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_OursGlobal_train.pickle', 'wb') as handle:
pickle.dump(results_ours_global_train, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_OursGlobal_trainAnn.pickle', 'wb') as handle:
pickle.dump(results_ours_global_trainA, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_OursGlobal_test.pickle', 'wb') as handle:
pickle.dump(results_ours_global_test, handle, protocol=pickle.HIGHEST_PROTOCOL)
with open('synthetic_OursGlobal_testAux.pickle', 'wb') as handle:
pickle.dump(results_ours_global_testA, handle, protocol=pickle.HIGHEST_PROTOCOL)
print("Execution done in %f mins"%((time.time()-start_time_exec)/60.))