CODEWITHSUNDEEP
pythonkeraslstm
robinster
robinster

Reputation: 21

!! ValueError: Error when checking input: expected lstm_2_input to have 3 dimensions, but got array with shape (4982, 12)

I am new in Deep Learning, and I'm trying to create this simple LSTM architecture in Keras using Google Colab:

  1. Input layer of 12 input neurons
  2. One Recurrent hidden layer of 1 hidden neuron for now
  3. Output layer of 1 output neuron

The original error was: 

ValueError: Error when checking input: expected lstm_2_input to have 3 dimensions, but got array with shape (4982, 12).

Then I tried:

input_shape=train_x.shape[1:]

But I got:

ValueError: Input 0 is incompatible with layer lstm_1: expected ndim=3, found ndim=2

Then I tried:

X_train = np.reshape(X_train, X_train.shape + (1,))

But I got another error again:

ValueError: Must pass 2-d input

Then I tried:

train_x = np.reshape(train_x, (train_x.shape[0], 1, train_x.shape[1]))

But it didn't work:

Must pass 2-d input

Here is my original code:

df_tea = pd.read_excel('cleaned2Apr2019pt2.xlsx')
df_tea.head()

train_x, valid_x = model_selection.train_test_split(df_tea,random_state=2, stratify=df_tea['offer_Offer'])

train_x.shape #(4982, 12)
valid_x.shape #(1661, 12)

model = Sequential()
model.add(LSTM(32, input_shape=train_x.shape, return_sequences=True))
model.add(LSTM(32, return_sequences=True))
model.add(Dense(1, activation='sigmoid'))

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['acc'])
history = model.fit(train_x, valid_x,
                    epochs=10,
                    batch_size=128,
                    validation_split=0.2)

I have looked through several stackoverflow and github suggestions for a similar problem, but none works.

Could someone help me please as I don't understand why all these methods failed.

Upvotes: 1

Views: 1844

Answers (1)

user44875
user44875

Reputation: 396

According to your code, timesteps = 1 (in LSTM terminology), input_dim = 12. Hence you should let input_shape = (1,12) A general formula is input_shape = (None, timesteps, input_dim) or input_shape = (timesteps, input_dim)

An example:

import numpy as np
from keras.layers import LSTM, Dense
from keras.models import Sequential
n_examples = 4982 #number of examples
n_ft = 12 #number of features
train_x= np.random.randn(n_examples, n_ft)
#valid_x.shape #(1661, 12)

model = Sequential()
model.add(LSTM(32, input_shape=(1, n_ft), return_sequences=True))
model.add(LSTM(32, return_sequences=True))
model.add(Dense(1, activation='sigmoid'))

model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['acc'])
model.summary()

Upvotes: 2

Related Questions