我做出來的核心learning.I都創建一個大小3的n_gram模型我用sklearn的HashingVectorizer這個purpose.Then我不得不使用keras建立神經網絡。但是我不知道怎麼養活輸入形狀n_gram模型HashingVectorizer,並使用它與keras
vec = HashingVectorizer(decode_error = 'ignore', n_features = 2**20, ngram_range = (3,3))
X = vec.fit_transform(tags)
y = np_utils.to_categorical(tag)
print(X.shape)
print(y.shape)
model = Sequential()
model.add(Dense(1024, input_shape = (1,X.shape[1]), activation = 'softmax'))
model.add(Dense(y.shape[1], activation = 'softmax'))
model.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])
model.fit(X, y, epochs = 10, batch_size = 200)
我的第二個編輯: 這裏幾乎每一點是相同的還是它拋出錯誤。 的代碼是:
from sklearn.feature_extraction.text import HashingVectorizer
from sklearn.neural_network import MLPClassifier
from sklearn.feature_extraction import FeatureHasher
from sklearn.preprocessing import OneHotEncoder, LabelEncoder
from keras.models import Sequential
from keras.layers.recurrent import LSTM, GRU, SimpleRNN
from keras.layers import Dense
from keras.utils import np_utils
from numpy import array
from scipy.sparse import csr_matrix
text = open('eng.train').read().lower().split()
X_train = []
y_train = []
for i in range (len(text)):
if i % 4 == 0:
X_train.append(text[i])
if i % 4 == 1:
y_train.append(text[i])
unq_tags = []
for i in range(len(y_train)):
if y_train[i] not in unq_tags:
unq_tags.append(y_train[i])
#hashing X_train
vec = HashingVectorizer(decode_error = 'ignore', n_features = 2**15)
X = vec.fit_transform(X_train)
X.toarray()
#one hot encoding y_train
values = array(y_train)
label_encoder = LabelEncoder()
integer_encoded = label_encoder.fit_transform(values)
encoded = np_utils.to_categorical(integer_encoded)
print(type(X))
print(X.shape)
print(type(encoded))
print(encoded.shape)
model = Sequential()
model.add(SimpleRNN(1024, input_shape = (X.shape[1],), activation = 'softmax'))
model.add(Dense(y.shape[1], activation = 'softmax'))
model.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])
model.fit(X, y, epochs = 20, batch_size = 200)
它拋出的錯誤如下:
class 'scipy.sparse.csr.csr_matrix'>
(204567, 32768)
<class 'numpy.ndarray'>
(204567, 46)
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
<ipython-input-58-4ac701f1ade4> in <module>()
47 '''
48 model = Sequential()
---> 49 model.add(SimpleRNN(1024, input_shape = (X.shape[1],), activation = 'softmax'))
50 model.add(Dense(y.shape[1], activation = 'softmax'))
51 model.compile(loss = 'categorical_crossentropy', optimizer = 'adam', metrics = ['accuracy'])
/home/manish/anaconda3/lib/python3.6/site-packages/keras/models.py in add(self, layer)
420 # and create the node connecting the current layer
421 # to the input layer we just created.
--> 422 layer(x)
423
424 if len(layer.inbound_nodes) != 1:
/home/manish/anaconda3/lib/python3.6/site-packages/keras/layers/recurrent.py in __call__(self, inputs, initial_state, **kwargs)
250 else:
251 kwargs['initial_state'] = initial_state
--> 252 return super(Recurrent, self).__call__(inputs, **kwargs)
253
254 def call(self, inputs, mask=None, initial_state=None, training=None
):
/home/manish/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py in __call__(self, inputs, **kwargs)
509 # Raise exceptions in case the input is not compatible
510 # with the input_spec specified in the layer constructor.
--> 511 self.assert_input_compatibility(inputs)
512
513 # Collect input shapes to build layer.
home/manish/anaconda3/lib/python3.6/site-packages/keras/engine/topology.py in assert_input_compatibility(self, inputs)
411 self.name + ': expected ndim=' +
412 str(spec.ndim) + ', found ndim=' +
--> 413 str(K.ndim(x)))
414 if spec.max_ndim is not None:
415 ndim = K.ndim(x)
ValueError: Input 0 is incompatible with layer simple_rnn_8: expected ndim=3, found ndim=2
我僅取得了一些調整但如果指定input_shape = (1,X.shape[1])
,模型我得到這個錯誤
嘗試'input_shape =(X.shape [1],)'。還發布完整的堆棧跟蹤錯誤。 –
此錯誤某種程度上與RNN的內部狀態。 [文檔](https://keras.io/layers/recurrent/)指出input_shape應該是一個'三維張量與形狀(的batch_size,時間步長,input_dim)'。您可以省略'batch_size'並將'None'用於'timesteps'。所以在這種情況下,'嘗試input_shape =(無,X.shape [1])' –