python - keras LSTM 以正确的形状输入输入

Question

我从具有以下形状的 Pandas 数据框中获取一些数据

df.head()
>>>
Value USD   Drop 7  Up 7    Mean Change 7   Change      Predict
0.06480     2.0     4.0     -0.000429       -0.00420    4
0.06900     1.0     5.0     0.000274        0.00403     2
0.06497     1.0     5.0     0.000229        0.00007     2
0.06490     1.0     5.0     0.000514        0.00200     2
0.06290     2.0     4.0     0.000229        -0.00050    3

前 5 列是 X并预测 y .这就是我为模型预处理数据的方式

from keras.models import Sequential
from keras.callbacks import EarlyStopping
from keras.callbacks import ModelCheckpoint
import pandas as pd
from sklearn.model_selection import train_test_split
import numpy as np
from sklearn.metrics import accuracy_score
from keras.layers import LSTM
from sklearn import preprocessing

# Convert a Pandas dataframe to the x,y inputs that TensorFlow needs
def to_xy(df, target):
    result = []
    for x in df.columns:
        if x != target:
            result.append(x)
    # find out the type of the target column.  Is it really this hard? :(
    target_type = df[target].dtypes
    target_type = target_type[0] if hasattr(target_type, '__iter__') else target_type
    # Encode to int for classification, float otherwise. TensorFlow likes 32 bits.
    if target_type in (np.int64, np.int32):
        # Classification
        dummies = pd.get_dummies(df[target])
        return df.as_matrix(result).astype(np.float32), dummies.as_matrix().astype(np.float32)
    else:
        # Regression
        return df.as_matrix(result).astype(np.float32), df.as_matrix([target]).astype(np.float32)

# Encode text values to indexes(i.e. [1],[2],[3] for red,green,blue).
def encode_text_index(df, name):
    le = preprocessing.LabelEncoder()
    df[name] = le.fit_transform(df[name])
    return le.classes_

df['Predict'].value_counts()
>>>
4    1194
3     664
2     623
0     405
1      14
Name: Predict, dtype: int64

predictions = encode_text_index(df, "Predict")
predictions
>>>
array([0, 1, 2, 3, 4], dtype=int64)

X,y = to_xy(df,"Predict")
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, shuffle=False)

X_train
>>>
array([[ 6.4800002e-02,  2.0000000e+00,  4.0000000e+00, -4.2857142e-04,
        -4.1999999e-03],
       [ 6.8999998e-02,  1.0000000e+00,  5.0000000e+00,  2.7414286e-04,
         4.0300000e-03],
       [ 6.4970002e-02,  1.0000000e+00,  5.0000000e+00,  2.2857143e-04,
         7.0000002e-05],
       ...,
       [ 9.5987000e+02,  5.0000000e+00,  2.0000000e+00, -1.5831429e+01,
        -3.7849998e+01],
       [ 9.9771997e+02,  5.0000000e+00,  2.0000000e+00, -1.6948572e+01,
        -1.8250000e+01],
       [ 1.0159700e+03,  5.0000000e+00,  2.0000000e+00, -1.3252857e+01,
        -7.1700001e+00]], dtype=float32)

y_train
>>>
array([[0., 0., 0., 0., 1.],
       [0., 0., 1., 0., 0.],
       [0., 0., 1., 0., 0.],
       ...,
       [0., 0., 0., 0., 1.],
       [0., 0., 0., 0., 1.],
       [0., 0., 0., 0., 1.]], dtype=float32)

X_train[1]
>>>
array([6.8999998e-02, 1.0000000e+00, 5.0000000e+00, 2.7414286e-04,
       4.0300000e-03], dtype=float32)

X_train.shape
>>>
(2320, 5)

X_train[1].shape
>>>
(5,)

最后是 LSTM 模型(也可能看起来不是最好的编写方法，所以如果是这样的话，我也会欣赏内层的重写)

model = Sequential()
#model.add(LSTM(128, dropout=0.2, recurrent_dropout=0.2, input_shape=(None, 1)))
model.add(LSTM(50, dropout=0.2, return_sequences=True, input_shape=X_train.shape))
model.add(LSTM(50, dropout=0.2, return_sequences=True))
model.add(LSTM(50, dropout=0.2, return_sequences=True))
model.add(LSTM(50, dropout=0.2, return_sequences=True))
#model.add(Dense(50, activation='relu'))
model.add(Dense(y_train.shape[1], activation='softmax'))

#model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
#model.fit(X_train, y_train, epochs=1000)

model.compile(loss='categorical_crossentropy', optimizer='adam')
monitor = EarlyStopping(monitor='val_loss', min_delta=1e-2, patience=15, verbose=1, mode='auto')
checkpointer = ModelCheckpoint(filepath="best_weights.hdf5", verbose=0, save_best_only=True) # save best model

model.fit(X_train, y_train, validation_data=(X_test, y_test), callbacks=[monitor,checkpointer], verbose=2, epochs=1000)
model.load_weights('best_weights.hdf5') # load weights from best model

运行这个会抛出这个错误

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-67-a17835a382f6> in <module>()
     15 checkpointer = ModelCheckpoint(filepath="best_weights.hdf5", verbose=0, save_best_only=True) # save best model
     16 
---> 17 model.fit(X_train, y_train, validation_data=(X_test, y_test), callbacks=[monitor,checkpointer], verbose=2, epochs=1000)
     18 model.load_weights('best_weights.hdf5') # load weights from best model

c:\users\samuel\appdata\local\programs\python\python35\lib\site-packages\keras\engine\training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)
    948             sample_weight=sample_weight,
    949             class_weight=class_weight,
--> 950             batch_size=batch_size)
    951         # Prepare validation data.
    952         do_validation = False

c:\users\samuel\appdata\local\programs\python\python35\lib\site-packages\keras\engine\training.py in _standardize_user_data(self, x, y, sample_weight, class_weight, check_array_lengths, batch_size)
    747             feed_input_shapes,
    748             check_batch_axis=False,  # Don't enforce the batch size.
--> 749             exception_prefix='input')
    750 
    751         if y is not None:

c:\users\samuel\appdata\local\programs\python\python35\lib\site-packages\keras\engine\training_utils.py in standardize_input_data(data, names, shapes, check_batch_axis, exception_prefix)
    125                         ': expected ' + names[i] + ' to have ' +
    126                         str(len(shape)) + ' dimensions, but got array '
--> 127                         'with shape ' + str(data_shape))
    128                 if not check_batch_axis:
    129                     data_shape = data_shape[1:]

ValueError: Error when checking input: expected lstm_48_input to have 3 dimensions, but got array with shape (2320, 5)

我尝试了很多 X_train 输入形状的变体，但每一个都会引发一些错误，我还检查了 Keras docs但尚不清楚应如何将数据输入模型

建议中的第 1 次尝试

首先是 reshape X_train

data = np.resize(X_train,(X_train.shape[0],1,X_train.shape[1]))
model.add(LSTM(50, dropout=0.2, return_sequences=True, input_shape=data.shape))

这失败并出现错误

ValueError: Input 0 is incompatible with layer lstm_52: expected ndim=3, found ndim=4 

建议我将其输入为

model.add(LSTM(50, dropout=0.2, return_sequences=True, input_shape=X_train.shape[1:]))

抛出相同的错误

ValueError: Input 0 is incompatible with layer lstm_63: expected ndim=3, found ndim=2

建议二

使用 pandas 的默认 X,y

y = df['Predict']
X = df[['Value USD', 'Drop 7', 'Up 7', 'Mean Change 7', 'Change']]

X = np.array(X)
y = np.array(y)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, shuffle=False)

LSTM 也期望通过以下方式输入 (batch_size, timesteps, input_dim)
所以我尝试了这个

model.add(LSTM(50, dropout=0.2, return_sequences=True, input_shape=(100, 100, X_train.shape)))

引发此错误

TypeError: Error converting shape to a TensorShape: int() argument must be a string, a bytes-like object or a number, not 'tuple'.

和不同的方式

model.add(LSTM(50, dropout=0.2, return_sequences=True, input_shape=(100, 100, X_train[1].shape)))

返回相同的错误

TypeError: Error converting shape to a TensorShape: int() argument must be a string, a bytes-like object or a number, not 'tuple'.

Answer 1

您想设置具有多个特征的 LSTM(有状态还是无状态？)，特征是列 Value USD Drop 7 Up 7 Mean Change 7 Change在您的数据框中。 https://github.com/keras-team/keras/issues/6471 中也有类似的问题。

Keras LSTM 接受输入为 (batch_size (number of samples processed at a time),timesteps,features) = (batch_size, timesteps, input_dim)因为您有 5 个功能 input_dim = features = 5 .我不知道你的全部数据，所以我不能说更多。 number_of_samples的关系(数据框中的行数)和 batch_size在 http://philipperemy.github.io/keras-stateful-lstm/ , batch_size是一次处理的样本数(行)(doubts regarding batch size and time steps in RNN):

Said differently, whenever you train or test your LSTM, you first have to build your input matrix X of shape nb_samples, timesteps, input_dim where your batch size divides nb_samples. For instance, if nb_samples=1024 and batch_size=64, it means that your model will receive blocks of 64 samples, compute each output (whatever the number of timesteps is for every sample), average the gradients and propagate it to update the parameters vector.

来源: http://philipperemy.github.io/keras-stateful-lstm/

批量大小对训练很重要

A batch size of 1 means that the model will be fit using online training (as opposed to batch training or mini-batch training). As a result, it is expected that the model fit will have some variance.

来源: https://machinelearningmastery.com/stateful-stateless-lstm-time-series-forecasting-python/
timesteps是您要回顾的时间步数/过去的网络状态，由于性能原因，LSTM 的最大值约为 200-500(梯度消失问题)，最大值约为 200( https://github.com/keras-team/keras/issues/2057)

拆分更容易( Selecting multiple columns in a pandas dataframe):

y = df['Predict']
X = df[['Value USD','Drop 7','Up 7','Mean Change 7', 'Change']]

在 https://www.kaggle.com/mknorps/titanic-with-decision-trees 是修改数据类型的代码

更新:

要摆脱这些错误，您必须像 中那样 reshape 训练数据。 Error when checking model input: expected lstm_1_input to have 3 dimensions, but got array with shape (339732, 29) (还包含超过 1 个时间步的 reshape 代码)。我发布了对我有用的整个代码，因为这个问题没有第一眼看上去那么简单( 注意[ 和] 的数量，它们在 reshape 时表示数组的维度):

import pandas as pd
import numpy as np

from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.callbacks import EarlyStopping
from keras.callbacks import ModelCheckpoint
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from keras.layers import LSTM
from sklearn import preprocessing

df = pd.read_csv('/path/data_lstm.dat')

y = df['Predict']
X = df[['Value USD', 'Drop 7', 'Up 7', 'Mean Change 7', 'Change']]

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, shuffle=False)

X_train_array = X_train.values  ( https://stackoverflow.com/questions/13187778/convert-pandas-dataframe-to-numpy-array-preserving-index )
y_train_array = y_train.values.reshape(4,1)

X_test_array = X_test.values
y_test_array = y_test.values


# reshaping to fit batch_input_shape=(4,1,5) batch_size, timesteps, number_of_features , batch_size can be varied batch_input_shape=(2,1,5), = (1,1,5),... is also working

X_train_array = np.reshape(X_train_array, (X_train_array.shape[0], 1, X_train_array.shape[1]))
#>>> X_train_array    NOTE THE NUMBER OF [ and ] !!
#array([[[ 6.480e-02,  2.000e+00,  4.000e+00, -4.290e-04, -4.200e-03]],

#       [[ 6.900e-02,  1.000e+00,  5.000e+00,  2.740e-04,  4.030e-03]],

#       [[ 6.497e-02,  1.000e+00,  5.000e+00,  2.290e-04,  7.000e-05]],

#       [[ 6.490e-02,  1.000e+00,  5.000e+00,  5.140e-04,  2.000e-03]]])
y_train_array = np.reshape(y_train_array, (y_train_array.shape[0], 1, y_train_array.shape[1]))
#>>> y_train_array     NOTE THE NUMBER OF [ and ]   !!
#array([[[4]],

#       [[2]],

#       [[2]],

#       [[2]]])



model = Sequential()
model.add(LSTM(32, return_sequences=True, batch_input_shape=(4,1,5) ))
model.add(LSTM(32, return_sequences=True ))
model.add(Dense(1, activation='softmax'))
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])