histnet/imdb_experiments.py at master · pglez82/histnet · GitHub

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
from ann.dqn import DQNNet
import sys,os
import pandas as pd
import numpy as np
import torch
from tensorflow.keras.preprocessing import sequence
from ann.histnet.histnet import HistNet
from ann.utils import BagGenerator
from ann.histnet.featureextraction.lstm import LSTMFeatureExtractionModule
from quantification.utils import absolute_error,relative_absolute_error,binary_kld
from datetime import datetime
from tqdm import tqdm

#Quapy imports. Until the library is stable enough, lets depend from our local clone
sys.path.append(r'/media/nas/pgonzalez/QuaPy')
import quapy as qp
from quapy.method.meta import QuaNet
from quapy.classification.neural import NeuralClassifierTrainer, LSTMnet


vocabulary_size = 5000
embedding_size = 150
max_len = 200
dropout = 0.5
sample_size=500
device = torch.device('cuda:0')
seed_value=2032

np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False

dataset_name='imdb'

dataset = qp.datasets.fetch_reviews(dataset_name,pickle=True)
qp.data.preprocessing.index(dataset, inplace=True,max_features=5000)

vocabulary_size = dataset.vocabulary_size

x_train, y_train = dataset.training.instances, dataset.training.labels
x_test, y_test = dataset.test.instances, dataset.test.labels

x_train = sequence.pad_sequences(x_train, maxlen=max_len)
x_test = sequence.pad_sequences(x_test, maxlen=max_len)

datatrain = qp.data.LabelledCollection(x_train, y_train)

#HistNet
fe = LSTMFeatureExtractionModule(vocabulary_size=vocabulary_size,embedding_size=embedding_size,
                                hidden_size=128,
                                output_size=32,dropout_lstm=0,dropout_linear=0.2)
histnetc = HistNet(train_epochs=5000,test_epochs=1,start_lr=0.001,end_lr=0.001,n_bags=500,bag_size=sample_size,n_bins=8,random_seed = seed_value,linear_sizes=[128],
                    feature_extraction_module=fe,bag_generator=BagGenerator(device=device),batch_size=16,loss_function=torch.nn.L1Loss(),
                    dropout=0.3,epsilon=0.005,weight_decay=0.00005,histogram='softrbf',use_labels=True,use_labels_epochs=10,val_split=0.2,device=device,verbose=1,dataset_name=dataset_name)
histnetc.fit(X=x_train, y=y_train)

np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)

#HistNet No Labels
fe = LSTMFeatureExtractionModule(vocabulary_size=vocabulary_size,embedding_size=embedding_size,
                                hidden_size=128,
                                output_size=32,dropout_lstm=0,dropout_linear=0.2)
histnet = HistNet(train_epochs=5000,test_epochs=1,start_lr=0.001,end_lr=0.001,n_bags=500,bag_size=sample_size,n_bins=8,random_seed = seed_value,linear_sizes=[128],
                    feature_extraction_module=fe,bag_generator=BagGenerator(device=device),batch_size=16,loss_function=torch.nn.L1Loss(),
                    dropout=0.3,epsilon=0.005,weight_decay=0.00005,histogram='softrbf',use_labels=False,val_split=0.2,device=device,verbose=1,dataset_name=dataset_name)
histnet.fit(X=x_train, y=y_train)

np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)


#Basic classifiers
np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)

classifier_basic = LSTMnet(vocabulary_size=vocabulary_size,n_classes=2,embedding_size=embedding_size,hidden_size=128,repr_size=100,drop_p=dropout)
learner_basic = NeuralClassifierTrainer(classifier_basic,padding_length=max_len,device='cuda:0')
learner_basic.fit(datatrain.instances,datatrain.labels)

cc_quapy = qp.method.aggregative.CC(learner_basic).fit(datatrain,fit_learner=False)
pcc_quapy = qp.method.aggregative.PCC(learner_basic).fit(datatrain,fit_learner=False)
ac_quapy = qp.method.aggregative.ACC(learner_basic).fit(datatrain,fit_learner=False)
pac_quapy = qp.method.aggregative.PACC(learner_basic).fit(datatrain,fit_learner=False)
hdy_quapy = qp.method.aggregative.HDy(learner_basic).fit(datatrain,fit_learner=False)

#train a classifier for the data
classifier_quanet = LSTMnet(vocabulary_size=vocabulary_size,n_classes=2,embedding_size=embedding_size,hidden_size=128,repr_size=100,drop_p=dropout)
learner_quanet = NeuralClassifierTrainer(classifier_quanet,padding_length=max_len,device='cuda:0')

#Fit quanet
np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)

quanet = QuaNet(learner_quanet,checkpointname='QuaNet-IMDB.dat',sample_size=sample_size,device='cuda')
quanet.fit(datatrain)

np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)

#DQNUsing
fe = LSTMFeatureExtractionModule(vocabulary_size=vocabulary_size,embedding_size=embedding_size,
                                hidden_size=128,
                                output_size=32,dropout_lstm=0,dropout_linear=0.2)
dqn = DQNNet(train_epochs=5000,test_epochs=1,start_lr=0.001,end_lr=0.001,n_bags=500,bag_size=sample_size,random_seed = seed_value,linear_sizes=[128],
                    feature_extraction_module=fe,bag_generator=BagGenerator(device=device),batch_size=16,loss_function=torch.nn.L1Loss(),
                    dropout=0.3,epsilon=0.005,weight_decay=0.00005,val_split=0.2,device=device,verbose=1,dataset_name=dataset_name)
dqn.fit(X=x_train, y=y_train)

methods = [cc_quapy,pcc_quapy,ac_quapy,pac_quapy,hdy_quapy,quanet,dqn,histnet,histnetc]
method_names = ['cc_quapy','pcc_quapy','ac_quapy','pac_quapy','hdy_quapy','quanet','dqn','histnet','histnetc']

np.random.seed(seed=seed_value)
torch.manual_seed(seed_value)

results=pd.DataFrame(columns=(['true']+method_names),dtype='float')
results_mae = pd.DataFrame(columns=(method_names),dtype='float')
results_mrae = pd.DataFrame(columns=(method_names),dtype='float')
results_kld = pd.DataFrame(columns=(method_names),dtype='float')
datatest = qp.data.LabelledCollection(x_test, y_test)

print("Making predictions...")
for r in tqdm(range(2000)):
    p=np.random.random()*(0.99-0.01) + 0.01
    index = datatest.sampling_index(sample_size, p,1-p)
    sample = datatest.sampling_from_index(index)
    results.loc[r,'true']=p
    predictions_test = learner_basic.predict_proba(x_test).astype('float')
    for method,method_name in zip(methods,method_names):
        if isinstance(method,qp.method.base.BaseQuantifier):
            predictions = method.quantify(sample.instances)
        else:
            predictions = method.predict(sample.instances)
        mae = absolute_error(p, predictions[0])
        mrae = relative_absolute_error(p,predictions[0])
        kld = binary_kld(p,predictions[0])
        results.loc[r,method_name]=predictions[0]
        results_mae.loc[r,method_name]=mae
        results_mrae.loc[r,method_name]=mrae
        results_kld.loc[r,method_name]=kld

    now = datetime.now()

results_mae.loc['mean'] = results_mae.mean()
results_mrae.loc['mean'] = results_mrae.mean()
results_kld.loc['mean'] = results_kld.mean()

if not os.path.exists("resultsimdb"):
    os.makedirs("resultsimdb")

filename1 = 'resultsimdb/results_imdb_{}.csv'.format(now.strftime('%Y-%m-%d_%H-%M-%S'))
filename2 = 'resultsimdb/results_mae_imdb_{}.csv'.format(now.strftime('%Y-%m-%d_%H-%M-%S'))
filename3 = 'resultsimdb/results_mrae_imdb_{}.csv'.format(now.strftime('%Y-%m-%d_%H-%M-%S'))
filename4 = 'resultsimdb/results_kld_imdb_{}.csv'.format(now.strftime('%Y-%m-%d_%H-%M-%S'))


with open(filename1, 'a') as the_file:
    the_file.write(str(histnet.__dict__))

results.to_csv(filename1,float_format='%.5f',mode='a')
results_mae.to_csv(filename2,float_format='%.5f',mode='a')
results_mrae.to_csv(filename3,float_format='%.5f',mode='a')
results_kld.to_csv(filename4,float_format='%.5f',mode='a')