When I imported Pretrained Keras model into the Matlab windowing size don't remain same?

Hi,

I've trained the model with window size 50, and my model input is (num_window,50,5). I've saved the model .h5 extended and imported it into the MATLAB with importKerasNetwork func. But when i use the window size 50 in MATLAB, following error is raising

'Error using DAGNetwork/predict

The prediction sequences are of feature dimension 50 but the input layer expects sequences of feature dimension 5.'

What could be reason? Please help.

Answer 1

[Jaimin]

Hello @MUHAMMED FURKAN YILMAZ

Based on the model description, I have created a sample model using Keras (version 2.6.0), which is as follows.

from keras.models import Sequential
from keras.layers import LSTM, Dense

model = Sequential()
model.add(LSTM(64, input_shape=(50, 5), return_sequences=True))
model.add(LSTM(32))
model.add(Dense(1, activation='linear'))

model.compile(optimizer='adam', loss='mean_squared_error')

model.save('model.h5')

It seems the error occurred because the model expects input with dimensions (50, 5), but the input was provided with dimensions (num_window, 50, 5). I have found a workaround to resolve the given error. You can iterate through each window one by one, so the model input becomes (50, 5).

Here I have attach a code snippet for better understanding.

net = importKerasNetwork('model.h5');

% Define the dimensions
num_window = 100; % You can change this to the desired number of windows

% Generate random data
% This will create a 3D array with dimensions (num_window, 50, 5)
data = rand(num_window, 5, 50);

% Display the size of the generated data to confirm
disp(size(Data));

% Make predictions using the imported Keras model
for i=1:num_window
    predictions = predict(net, reshape(data(i,:,:),5,50))
end

I hope this will be helpful.