请　[注册]　或　[登录]　 | 返回主站

量化交易吧 /  数理科学 帖子：3364477 新帖：3

深度学习_分类：LSTM用于分类

交易资深人士发表于：9 月 16 日 21：35回复(1) 
import keras
from keras.models import Sequential
from keras.layers import LSTM, Dense, Dropout
import numpy as np

/Users/jiaohaibin/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Using TensorFlow backend.

data = 'abcdefghijklmnopqrstuvwxyz'
#data_set = set(data)
data_set = list(data) #使用列表
word_len = len(data_set) #26

#制作字典
word_2_int = {b:a for a,b in enumerate(data_set)}

#交换位置
int_2_word = {a:b for a,b in enumerate(data_set)}

print(word_2_int)
print(int_2_word)
word_len

{'a': 0, 'b': 1, 'c': 2, 'd': 3, 'e': 4, 'f': 5, 'g': 6, 'h': 7, 'i': 8, 'j': 9, 'k': 10, 'l': 11, 'm': 12, 'n': 13, 'o': 14, 'p': 15, 'q': 16, 'r': 17, 's': 18, 't': 19, 'u': 20, 'v': 21, 'w': 22, 'x': 23, 'y': 24, 'z': 25}
{0: 'a', 1: 'b', 2: 'c', 3: 'd', 4: 'e', 5: 'f', 6: 'g', 7: 'h', 8: 'i', 9: 'j', 10: 'k', 11: 'l', 12: 'm', 13: 'n', 14: 'o', 15: 'p', 16: 'q', 17: 'r', 18: 's', 19: 't', 20: 'u', 21: 'v', 22: 'w', 23: 'x', 24: 'y', 25: 'z'}

26
def words_2_ints(words):
    ints = []
    for itmp in words:
        ints.append(word_2_int[itmp])
    return ints
 
print(words_2_ints('ab'))
 
def words_2_one_hot(words, num_classes=word_len):
    return keras.utils.to_categorical(words_2_ints(words), num_classes=num_classes)
print(words_2_one_hot('a'))

[0, 1]
[[1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
  0. 0.]]

def get_one_hot_max_idx(one_hot):
    idx_ = 0
    max_ = 0
    for i in range(len(one_hot)):
        if max_ < one_hot[i]:
            max_ = one_hot[i]
            idx_ = i
    return idx_
 
def one_hot_2_words(one_hot):
    tmp = []
    for itmp in one_hot:
        tmp.append(int_2_word[get_one_hot_max_idx(itmp)])
    return "".join(tmp)
 
words_2_one_hot('abcd')[0]

array([1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
       0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)
print( one_hot_2_words(words_2_one_hot('abcd')) )

abcd

time_step = 3 #一个句子有3个词,句子的长度
 
def genarate_data(batch_size=5, genarate_num=100):
    #genarate_num = -1 表示一直循环下去,genarate_num=1表示生成一个batch的数据，以此类推
    #这里，我也不知道数据有多少，就这么循环的生成下去吧。
    #入参batch_size 控制一个batch 有多少数据，也就是一次要yield进多少个batch_size的数据
    '''
    例如，一个batch有batch_size=5个样本，那么对于这个例子，需要yield进的数据为：
    abc->d
    bcd->e
    cde->f
    def->g
    efg->h
    
    然后，把这些数据都转换成one-hot形式，最终数据，输入x的形式为：
    
    [第1个batch]
    [第2个batch]
    ...
    [第genarate_num个batch]
    
    每个batch的形式为：句子组成的列表
    
    [第1句话（如abc）]
    [第2句话（如bcd）]
    ...
    
    
    每一句话的形式为：one-hot词向量组成的列表
    
    [第1个词的one-hot表示]
    [第2个词的one-hot表示]
    ...
    
    '''
    cnt = 0
    batch_x = []
    batch_y = []
    sample_num = 0
    while(True):
        for i in range(len(data) - time_step):
            batch_x.append(words_2_one_hot(data[i : i+time_step]))
            batch_y.append(words_2_one_hot(data[i+time_step])[0]) 
            #这里数据加[0]，是为了符合keras的输出数据格式。
            #因为不加[0]，表示是3维的数据。 你可以自己尝试不加0，看下面的test打印出来是什么
            sample_num += 1
            #print('sample num is :', sample_num)
            if len(batch_x) == batch_size:
                yield (np.array(batch_x), np.array(batch_y))
                batch_x = []
                batch_y = []
                if genarate_num != -1:
                    cnt += 1
 
                if cnt == genarate_num:
                    return
            
for test in genarate_data(batch_size=3, genarate_num=1):
    print('--------x:')
    print(test[0])
    print('--------y:')
    print(test[0])

--------x:
[[[1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]]

 [[0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]]

 [[0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]]]
--------y:
[[[1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]]

 [[0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]]

 [[0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]
  [0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
   0. 0. 0.]]]

 

model = Sequential()
 
# LSTM输出维度为 128
# input_shape 控制输入数据的形态：

# time_stemp 表示一句话 有多少个单词 序列长度 为3个字母
# word_len  表示一个单词用多少维度表示，这里是26维
 
model.add(LSTM(128, input_shape=(time_step, word_len))) # 3*26
model.add(Dense(word_len, activation='softmax')) 
#输出用一个softmax，来分类，维度就是26，预测是哪一个字母,26个字母
 
model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'] )
model.summary()
#print(model.summary())

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_5 (LSTM)                (None, 128)               79360     
_________________________________________________________________
dense_5 (Dense)              (None, 26)                3354      
=================================================================
Total params: 82,714
Trainable params: 82,714
Non-trainable params: 0
_________________________________________________________________

history =model.fit_generator(generator=genarate_data(batch_size=5, genarate_num=-1),
                             epochs=50, steps_per_epoch=10)

#steps_per_epoch的意思是，一个epoch中，执行多少个batch
#batch_size 样本个数，在一个batch中，有多少个样本。，
#所以，batch_size*steps_per_epoch就等于一个epoch中，训练的样本数量。
#(这个说法不对！再观察看看吧)
#可以将epochs设置成1，或者2，然后在genarate_data中打印样本序号，观察到样本总数。

Epoch 1/50
10/10 [==============================] - 1s 139ms/step - loss: 3.2392 - acc: 0.1400
Epoch 2/50
10/10 [==============================] - 0s 9ms/step - loss: 3.1828 - acc: 0.4400
Epoch 3/50
10/10 [==============================] - 0s 8ms/step - loss: 3.1374 - acc: 0.7400
Epoch 4/50
10/10 [==============================] - 0s 9ms/step - loss: 3.0891 - acc: 0.8400
Epoch 5/50
10/10 [==============================] - 0s 8ms/step - loss: 3.0287 - acc: 0.9200
Epoch 6/50
10/10 [==============================] - 0s 7ms/step - loss: 2.9627 - acc: 0.9600
Epoch 7/50
10/10 [==============================] - 0s 7ms/step - loss: 2.8829 - acc: 0.9800
Epoch 8/50
10/10 [==============================] - 0s 7ms/step - loss: 2.7913 - acc: 1.0000
Epoch 9/50
10/10 [==============================] - 0s 7ms/step - loss: 2.6982 - acc: 1.0000
Epoch 10/50
10/10 [==============================] - 0s 7ms/step - loss: 2.5757 - acc: 1.0000
Epoch 11/50
10/10 [==============================] - 0s 7ms/step - loss: 2.4265 - acc: 1.0000
Epoch 12/50
10/10 [==============================] - 0s 7ms/step - loss: 2.2556 - acc: 1.0000
Epoch 13/50
10/10 [==============================] - 0s 7ms/step - loss: 2.0592 - acc: 1.0000
Epoch 14/50
10/10 [==============================] - 0s 6ms/step - loss: 1.8639 - acc: 1.0000
Epoch 15/50
10/10 [==============================] - 0s 6ms/step - loss: 1.6798 - acc: 1.0000
Epoch 16/50
10/10 [==============================] - 0s 6ms/step - loss: 1.4371 - acc: 1.0000
Epoch 17/50
10/10 [==============================] - 0s 7ms/step - loss: 1.2097 - acc: 1.0000
Epoch 18/50
10/10 [==============================] - 0s 6ms/step - loss: 0.9973 - acc: 1.0000
Epoch 19/50
10/10 [==============================] - 0s 6ms/step - loss: 0.8093 - acc: 1.0000
Epoch 20/50
10/10 [==============================] - 0s 6ms/step - loss: 0.6408 - acc: 1.0000
Epoch 21/50
10/10 [==============================] - 0s 6ms/step - loss: 0.5214 - acc: 1.0000
Epoch 22/50
10/10 [==============================] - 0s 6ms/step - loss: 0.3904 - acc: 1.0000
Epoch 23/50
10/10 [==============================] - 0s 6ms/step - loss: 0.2961 - acc: 1.0000
Epoch 24/50
10/10 [==============================] - 0s 5ms/step - loss: 0.2180 - acc: 1.0000
Epoch 25/50
10/10 [==============================] - 0s 6ms/step - loss: 0.1584 - acc: 1.0000
Epoch 26/50
10/10 [==============================] - 0s 6ms/step - loss: 0.1140 - acc: 1.0000
Epoch 27/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0888 - acc: 1.0000
Epoch 28/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0648 - acc: 1.0000
Epoch 29/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0490 - acc: 1.0000
Epoch 30/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0371 - acc: 1.0000
Epoch 31/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0268 - acc: 1.0000
Epoch 32/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0197 - acc: 1.0000
Epoch 33/50
10/10 [==============================] - 0s 6ms/step - loss: 0.0151 - acc: 1.0000
Epoch 34/50
10/10 [==============================] - 0s 10ms/step - loss: 0.0111 - acc: 1.0000
Epoch 35/50
10/10 [==============================] - 0s 10ms/step - loss: 0.0081 - acc: 1.0000
Epoch 36/50
10/10 [==============================] - 0s 8ms/step - loss: 0.0060 - acc: 1.0000
Epoch 37/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0042 - acc: 1.0000
Epoch 38/50
10/10 [==============================] - 0s 6ms/step - loss: 0.0031 - acc: 1.0000
Epoch 39/50
10/10 [==============================] - 0s 6ms/step - loss: 0.0023 - acc: 1.0000
Epoch 40/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0016 - acc: 1.0000
Epoch 41/50
10/10 [==============================] - 0s 7ms/step - loss: 0.0012 - acc: 1.0000
Epoch 42/50
10/10 [==============================] - 0s 10ms/step - loss: 8.4960e-04 - acc: 1.0000
Epoch 43/50
10/10 [==============================] - 0s 11ms/step - loss: 5.8763e-04 - acc: 1.0000
Epoch 44/50
10/10 [==============================] - 0s 9ms/step - loss: 4.2971e-04 - acc: 1.0000
Epoch 45/50
10/10 [==============================] - 0s 8ms/step - loss: 3.1050e-04 - acc: 1.0000
Epoch 46/50
10/10 [==============================] - 0s 8ms/step - loss: 2.1903e-04 - acc: 1.0000
Epoch 47/50
10/10 [==============================] - 0s 12ms/step - loss: 1.5835e-04 - acc: 1.0000
Epoch 48/50
10/10 [==============================] - 0s 11ms/step - loss: 1.1565e-04 - acc: 1.0000
Epoch 49/50
10/10 [==============================] - 0s 10ms/step - loss: 8.0106e-05 - acc: 1.0000
Epoch 50/50
10/10 [==============================] - 0s 11ms/step - loss: 6.0480e-05 - acc: 1.0000

history.history['acc']

[0.14000000208616256,
 0.44000001102685926,
 0.7400000095367432,
 0.8400000095367431,
 0.9200000047683716,
 0.9600000023841858,
 0.9800000011920929,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0,
 1.0]
import matplotlib.pyplot as plt
epochs = range(len(acc)) # 横坐标的长度
plt.figure()
acc = history.history['acc']
#val_acc = history.history['val_acc']
loss = history.history['loss']
#val_loss = history.history['val_loss']
#线条
plt.plot(epochs, acc, 'bo', label='Training acc')
#plt.plot(epochs, val_acc, 'b', label='Validation acc')

plt.title('Training and validation accuracy')#标题
plt.legend() #角标

plt.show()

plt.figure()
#线条
plt.plot(epochs, loss, 'bo', label='Training loss')
#plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss') #标题
plt.legend()#角标

plt.show()

result = model.predict(np.array([words_2_one_hot('bcd')]))
print(one_hot_2_words(result))

e

       

全部回复

0/140

量化课程

    移动端课程