-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
150 lines (126 loc) · 6.12 KB
/
train.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
import numpy as np
import pandas as pd
import torch
from datasets import load_dataset
from transformers import AutoModel
from transformers import AutoTokenizer
import umap
from sklearn.preprocessing import MinMaxScaler
import matplotlib.pyplot as plt
from matplotlib import rcParams
# 设置字体为支持中文的字体
rcParams['font.sans-serif'] = ['Microsoft YaHei']
rcParams['axes.unicode_minus'] = False # 解决负号'-'显示为方块的问题
# pip install umap-learn
# pip install scikit-learn
from sklearn.linear_model import LogisticRegression
from sklearn.dummy import DummyClassifier
def print_type_structure(obj, indent=0):
if isinstance(obj, np.ndarray):
print(' ' * indent + f'ndarray.size(): {obj.size}')
elif isinstance(obj, torch.Tensor):
print(' ' * indent + f'Tensor: {obj.size()}')
elif isinstance(obj, (list, tuple)):
print(' ' * indent + 'List/Tuple:')
for item in obj:
print_type_structure(item, indent + 4)
elif isinstance(obj, dict):
print(' ' * indent + 'Dictionary:')
for key, value in obj.items():
print(' ' * (indent + 4) + f'Key: {type(key).__name__} # ->"'+key+'", Value:')
print_type_structure(value, indent + 8)
else:
print(' ' * indent + f'Type: {type(obj).__name__}')
def export_hidden_states(batch):
# inputs = {k:v.to(device)for k,v in batch.items()}
inputs = {k: v for k, v in batch.items()}
with torch.no_grad():
last_hidden_states = model(**inputs).last_hidden_state
return {"hidden_state":last_hidden_states[:,0].cpu().numpy()}
def tokenize_function(examples):
return tokenizer(examples["text"], padding="max_length", truncation=True)
def exec2(emotions):
emotions_encoded = emotions.map(tokenize_function, batched=True, batch_size=100)
print_type_structure(emotions_encoded)
print("^^^^^^^^^^^^")
emotions_encoded.set_format(type="torch", columns=["attention_mask","input_ids", "label"])
print_type_structure(emotions_encoded)
# emotions_encoded = emotions_encoded.select(["attention_mask","input_ids", "label"])
emotions_encoded=emotions_encoded.remove_columns(["text"])
tempLabel = emotions_encoded["label"]
emotions_encoded=emotions_encoded.remove_columns(["label"])
emotions_encoded = emotions_encoded.map(export_hidden_states, batched=True, batch_size=100)
emotions_encoded = emotions_encoded.add_column("label", emotions["label"])
print_type_structure(emotions_encoded)
print("emotions_encoded:" + str(emotions_encoded.column_names))
# emotions_encoded.add_column("label", tempLabel)
return emotions_encoded
def showHex(emotions_encoded):
x_train = np.array(emotions_encoded["hidden_state"])
x_scaled = MinMaxScaler().fit_transform(x_train)
xumap = umap.UMAP(n_components=2).fit(x_scaled)
df_train_umap = pd.DataFrame(xumap.embedding_, columns=["x", "y"])
df_train_umap["label"] = np.array(emotions_encoded["label"])
print(df_train_umap.head(10))
map_label = {0: '悲伤', 1: '喜悦', 2: '爱', 3: '愤怒', 4: '恐惧', 5: '惊喜'}
cmaps = ["Greens", "Blues", "Reds", "Purples", "Oranges", "Greys"]
fig, axes = plt.subplots(2, 3, figsize=(10, 10))
axes = axes.flatten()
print(emotions_encoded.features["label"])
for i, (label, cmap) in enumerate(zip(df_train_umap["label"].unique(), cmaps)):
print(f"label == {label},i={i},axes={axes}")
df_train_umap_label = df_train_umap.query(f"label == {label}")
axes[i].hexbin(df_train_umap_label["x"], df_train_umap_label["y"], label=label, gridsize=20,
linewidths=(0,), cmap=cmap)
axes[i].set_title(map_label[label])
plt.show()
def plot_confusion_matrix(y_preds,y_true,labels,title='Confusion matrix', cmap=plt.cm.Blues):
from sklearn.metrics import confusion_matrix,ConfusionMatrixDisplay
cm = confusion_matrix(y_true, y_preds,normalize="true")
fig,ax = plt.subplots(figsize=(8,8))
disp = ConfusionMatrixDisplay(confusion_matrix=cm,display_labels=labels)
disp.plot(include_values=True,cmap=cmap, ax=ax,values_format='.2f',xticks_rotation=45)
plt.title(title)
plt.show()
def main():
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f'GPU OR CPU: {device}')
emotions = load_dataset("emotion")
emotions_train = emotions["train"]
emotions_train = emotions_train.select(range(300))
print("@@@@@@@@@@@@@@@@@@@@@@")
trainout = exec2(emotions_train)
# break run
# showHex(trainout)
print("@@@@@@@@@@@@@@@@@@@@@@")
emotions_validation = emotions["validation"]
emotions_validation = emotions_validation.select(range(300))
validationout = exec2(emotions_validation)
# break run
# showHex(validationout)
x_train= np.array(trainout["hidden_state"])
x_val = np.array(validationout["hidden_state"])
y_train = np.array(trainout["label"])
y_val = np.array(validationout["label"])
print(x_train.shape,x_val.shape,y_train.shape,y_val.shape)
lr_clf = LogisticRegression()
# 使用训练数据拟合模型
lr_clf.fit(x_train,y_train)
# 方法默认计算的是分类准确率（accuracy）
print(lr_clf.score(x_val,y_val))
dummy_clf = DummyClassifier(strategy="most_frequent")
dummy_clf.fit(x_train,y_train)
print(dummy_clf.score(x_val,y_val))
map_label = {0: '悲伤', 1: '喜悦', 2: '爱', 3: '愤怒', 4: '恐惧', 5: '惊喜'}
# 训练结果
y_preds = lr_clf.predict(x_val)
#验证结果
# y_val
plot_confusion_matrix(y_preds,y_val,map_label.values(),title="Confusion matrix (normalized)",cmap=plt.cm.Greens)
model = AutoModel.from_pretrained("bert-base-uncased")
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
if __name__ == '__main__':
main()
# requests.exceptions.SSLError: (MaxRetryError(
# "HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /bert-base-uncased/resolve/main/config.json (Caused by SSLError(SSLEOFError(8, '[SSL: UNEXPECTED_EOF_WHILE_READING] EOF occurred in violation of protocol (_ssl.c:1006)')))"),
# '(Request ID: 3b533093-dbe1-4861-87fc-04bda776d007)')