How to solve ValueError in model.predict()? - tensorflow

Question

I am new in neural network problems. I have searched for couple of hours but could not understand what should I do to fix this issue! I'm working with nsl-kdd dataset for intrusion detection system with convolutional neural net.
I stuck with this problem : ValueError: Input 0 of layer dense_14 is incompatible with the layer: expected axis -1 of input shape to have value 3904 but received input with shape [None, 3712]
Shapes:
x_train (125973, 122)
y_train (125973, 5)
x_test (22544, 116)
y_test (22544,)
After reshape :
x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], 1)) #(125973, 122, 1)
x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], 1)) #(22544, 116, 1)
Model :
model = Sequential()
model.add(Convolution1D(64, 3, padding="same",activation="relu",input_shape = (x_train.shape[1], 1)))
model.add(MaxPooling1D(pool_size=(2)))
model.add(Flatten())
model.add(Dense(128, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(5, activation="softmax"))
Compile :
model.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])
model.fit(x_train, Y_train, epochs = 5, batch_size = 32)
pred = model.predict(x_test) #problem is occurring for this line
y_pred= np.argmax(pred, axis = 1)
model summary

Answer 1

Your x_test should have same dimensions as x_train.
x_train = (125973, 122, 1)
x_test = (22544, 116, 1) # the second parameter must match the train set
Code sample:
import tensorflow as tf
import pandas as pd
import numpy as np
from tensorflow.keras.layers import *
from tensorflow.keras import *
x1 = np.random.uniform(100, size =(125973, 122,1))
x2 = np.random.uniform(100, size =(22544, 122, 1))
y1 = np.random.randint(100, size =(125973,5), dtype = np.int32)
y2 = np.random.randint(2, size =(22544, ), dtype = np.int32)
def create_model2():
model = Sequential()
model.add(Convolution1D(64, 3, padding="same",activation="relu",input_shape = (x1.shape[1], 1)))
model.add(MaxPooling1D(pool_size=(2)))
model.add(Flatten())
model.add(Dense(128, activation="relu"))
model.add(Dropout(0.5))
model.add(Dense(5, activation="softmax"))
model.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])
return model
model = create_model2()
tf.keras.utils.plot_model(model, 'my_first_model.png', show_shapes=True)
You model looks like this:
Now if use your test set to create your model keeping your dimension as (22544, 116, 1).
You get a model that looks this.
As the dimensions are different the expected input and output of each layers are different.
When you have appropriate test dimensions the output works as expected:
pred = model.predict(x2)
pred
Output:
array([[1., 0., 0., 0., 0.],
[1., 0., 0., 0., 0.],
[1., 0., 0., 0., 0.],
...,
[1., 0., 0., 0., 0.],
[1., 0., 0., 0., 0.],
[1., 0., 0., 0., 0.]], dtype=float32)

Answer 2

Problem: The problem is that your test set does have the same dimensions as your training set. The test set should look as if you took a sample from your training set. So if your training set has the dimensions x_train.shape = (125973, 122) and y_train.shape = (125973, 5). Then your test set should have the dimensions x_test.shape = (sample_num, 122) and y_test.shape = (sample_num, 5).
Possible Solution: An easy way to do testing if you didn't want to use your test set with be with a validation split in the .fit().
So this: model.fit(x_train, Y_train, epochs = 5, batch_size = 32)
would turn into this: model.fit(x_train, Y_train, epochs = 5, batch_size = 32, validation_split=0.2)
This would chop off 20% of your training data and use that for testing. Then after every epoch, TensorFlow will print how the network performed on that validation data so that you can see how your model performs on data it has never seen before.