Train a neural network on the Covertype dataset. It needs to be defined arbitrary neural network architecture. Pay attention to the number of input and output neurons in the first and last layers of the neural network. Use an adequate error function. Train neural network arbitrary number of epochs with arbitrary batch size. Find out accuracy and response on the test data set before and after training. Define a validation data set. Implement an early stop during training so training is stopped when the error over the validation set starts to grow. Tolerate 2 epoch of error growth, and then stop the training if the error increases in the third as well consecutive epoch. Search the hyperparameter space. It is necessary to vary at least 2 values of the initial one training factors. In addition to the training factor, arbitrarily choose at least two other parameters which affects the network architecture and varies the given parameter by at least 2 values. Choose a model which gives the best precision on the validation set. Note: If only implement this point without the previous one, train all networks the same number of epochs. The use of standard libraries and all non-standard libraries is allowed. It is allowed also use libraries that are not used during the exercises (it is necessary to specify which libraries they are and include installation instructions within the project). The decision must be attached in kind Jupiter notebook file.
Train a neural network on the Covertype dataset. It needs to be defined arbitrary neural network architecture. Pay attention to the number of input and output neurons in the first and last layers of the neural network. Use an adequate error function. Train neural network arbitrary number of epochs with arbitrary batch size. Find out accuracy and response on the test data set before and after training. Define a validation data set. Implement an early stop during training so training is stopped when the error over the validation set starts to grow. Tolerate 2 epoch of error growth, and then stop the training if the error increases in the third as well consecutive epoch. Search the hyperparameter space. It is necessary to vary at least 2 values of the initial one training factors. In addition to the training factor, arbitrarily choose at least two other parameters which affects the network architecture and varies the given parameter by at least 2 values. Choose a model which gives the best precision on the validation set. Note: If only implement this point without the previous one, train all networks the same number of epochs. The use of standard libraries and all non-standard libraries is allowed. It is allowed also use libraries that are not used during the exercises (it is necessary to specify which libraries they are and include installation instructions within the project). The decision must be attached in kind Jupiter notebook file.
Step by step
Solved in 4 steps
