Neural Network Flow and Reference

This page supplies one of the neural network model flow and allocate the links for each page.

1.Data Preprocessing

In many cases, the data is not appropriate format for machine learning or deep learning algorithms.
So, we have to convert the data to the format that can be applied for learning algoritms.

2.Confirm the characteristics of the data(option)

It is useful information to confirm and visualize the characteristics of the data sometimes.

3.Model Definition

When we construct the neural network model, we think about the overfitting problem and data types and so on.
And, we might have to think sequential model is better? or functional model is better?
Sometimes sequential model is easy for writing models, sometimes the model we want to construct is limited to functional model.(functional model is more flexible.)

4.Learning

There are some types of learning method and some tuning method.
If you would like to the optimal hyperparameter, the number of the units or learning rate, it might be helpful to use the hyper parameter search.
And, in the case you use the dropout or some probalistic method, you have to change the mode when the prediction.

5.Evaluation of the model

Evaluation of your neural network model for checking the precision and recall in the clustering problem.
In the regression case, r2 score and rms(root mean squared error) is often used for evaluation.

The explanation above is the part of whole catalog page of ReNom, so if you like, it might be helpful to construct the model you have to make.