Apr 22, 2022
When you are done with the data preparation process, you can proceed to the model creation phase. Normally, Predibot creates the model with the basic parameters in seconds. You can directly press the "create" button without making any adjustments on model parameters.
This part is so easy with Predibot when your data is ready.
However, you can also make fine adjustments with the "Advanced Options" button before creating model. This action is called "parameter tuning" and can be configured to suit different data and task.
Iterations: The maximum tree number that can be created when learning process.
Learning Rate (shrinkage): It is for reducing the gradient step and affects the overall time of training. It can be in the range of 0 and 1. Using a low learning rate can dramatically improve the perfomance of your gradient boosting model. Usually a learning rate in the range of 0.1 to 0.3 gives the best results
Depth: It is depth of the tree using in the model. Depth optimally ranges from 4 to 10, from 6 to 9 are recommended.
One-hot Encoding Max Size: For categorical features, it sets distinct values less than or equal to the given parameter value.
Test Size: It is used to estimate the performance of model when they are used to make predictions on data not used to train the model.
Shuffle Data: If it is set, the order of rows changes randomly while splitting the data into test and training.
Use Best Model: Count of trees in model is defining automatically when this parameter is set. No trees are saved after this value.
Be careful when making these settings. Any wrong parameter can lead to overfitting, which means low error on train but again high error on validation and causes higher levels of error when applied to your datasets.
When your model is ready, you can always re-create it with different paremeters. When a learning process completed, you can analysis model information on "Model Info" tab.
Next: Model Info