Original subtitles

In this video, we will explain about dataset splitting.

The dataset must be split into two parts a training set and a test set.

Because this data is used for training, the training that has the most data following the completion

of the training process, the test set is used to evaluate the performance of the train model.

The data is split to prevent overfitting and to evaluate the model.

Overfitting occurs when a model performs well on a training set, but performs poorly on data that the

model has never seen before.

Commonly the size distribution of training and testing sets is 67% training and 33% testing 75% training

and 25% testing.

90% training and 10% testing.

The YOLO model has had two parameters to obtain optimal values.

These hyper parameters must be tuned during the training process.

Data is required to test the tuning hyper parameter, however, because the data used is not from the

training set an additional component, namely the validation set is required.

Typically the validation set is 10% to 20% of the training set.

We have provided a Python program for performing data set splitting.

The program will be explained in the section on training custom objects.

See you then.