Splitting a dataset into a training and a test set is a crucial step when building a deep learning model. The training set is used to train the model and the test set is used to evaluate the model’s performance on unseen data. In Keras, it’s also possible to split the training data into a training set and a validation set, this is called validation split.

Validation split allows to use a portion of the training data to validate the model’s performance during the training process. By doing this, it’s possible to detect overfitting early and make adjustments to the model before evaluating it on the test set.

In Keras, you can use the “validation_split” argument when calling the “fit” function on your model. This argument specifies the percentage of the training data that should be used as the validation set. For example, if you set the “validation_split” argument to 0.2, 20% of the training data will be used as the validation set and the remaining 80% will be used as the training set.

First, you need to have your input data and target data ready, and then you can pass them to the fit function along with the “validation_split” argument.

It’s important to note that when using “validation_split” in Keras, the data is split randomly. This means that the model could be trained and validated on different data every time it’s run, which is important to minimize the chances of overfitting.

Additionally, when calling the fit function, you can also specify the batch size and the number of epochs to train the model. The batch size is the number of samples per gradient update, and the number of epochs is the number of complete passes through the data.

This is a general description of how to split train and test datasets using validation_split in Keras, and is not an exhaustive guide.