In a genetic CNN, we use genetic algorithms to estimate the optimum CNN architecture; in genetic RNN, we will now use a genetic algorithm to find the optimum hyperparameters of the RNN, the window size, and the number of hidden units. We will find the parameters that reduce the root-mean-square error (RMSE) of the model.
The hyperparameters window size and number of units are again encoded in a binary string with 6 bits for window size and 4 bits for the number of units. Thus, the complete encoded chromosome will be of 10 bits. The LSTM is implemented using Keras.
The code we implement is taken from https://github.com/aqibsaeed/Genetic-Algorithm-RNN:
- The necessary modules are imported. This time, ...
