Comparing different versions - feature importance

I’ve run rasgo.evaluate.feature_importance multiple times after performing rounds of feature engineering, how can I compare the current version to a prior stage?

PyRasgo allows you to easily see each iteration in your feature engineering process through experiments. First, always begin the process by creating an experiment


Note that if you are performing other related work in another file or notebook, activating an experiment of the same name will link that work to the work done in the first file.

In the upper left corner, you can see which iteration of this experiment is currently being displayed.

Screen Shot 2021-06-14 at 11.03.04 AM

For example, this is iteration 19 of 19.

On the right hand side, you can click Experiment Leaderboard

Screen Shot 2021-06-14 at 11.03.13 AM

to see a list of all iterations of your experiment

You can sort these iterations by R^2, RMSE, number of features and time it was created.

Screen Shot 2021-06-14 at 11.04.12 AM

When you find the other iteration you would like to explore, click the blue Select button for that iteration

Screen Shot 2021-06-14 at 11.04.27 AM

Thanks! Can you create multiple experiments with different names?

Absolutely. You can create as many experiments as you like. Each time you run
rasgo.activate_experiment(‘Experiment Name’)
will attach all of your activities after that call to the experiment called Experiment Name.

This means that if you use the same experiment name in a different file or notebook, the feature profiles and importance you calculate will be attached to the experiment you had already defined.

To create a new experiment, simply create a new experiment with a new name
rasgo.activate_experiment(“New Experiment Name”)

You can see all of your experiments on by clicking the < Experiments in the upper left
Screen Shot 2021-06-14 at 11.48.51 AM

or click to open your profile in the upper right
Screen Shot 2021-06-14 at 12.01.29 PM
and select My Experiments
Screen Shot 2021-06-14 at 11.49.01 AM