Training Models#
The run Command#
To run the experiments, we can use the brisk run command. This will run the
experiments defined in the settings.py file using the specified workflow.
It is best to call the run command from the root of the project to ensure Brisk
can find the files.
The -w argument is used to specify the workflow file to use. This is the name
of the file in the workflows directory containing the workflow class, without
the .py extension. You can give the results directory a name using the -n
argument.
To run the workflow we just made you would call:
brisk run -w workflow -n tutorial_results
When you call this you will see a progress bar appear in the terminal. When all the experiments are complete, you will see a summary of the experiments run, with a status (PASSED or FAILED) and the time taken for each experiment.
You should see a summary table that looks like this:
======================================================================
EXPERIMENT SUMMARY
======================================================================
Group: tutorial
======================================================================
Dataset: diabetes
Experiment Status Time
----------------------------------------------------------------------
tutorial_linear PASSED 0m 1s
tutorial_ridge PASSED 0m 0s
tutorial_elasticnet PASSED 0m 0s
======================================================================
Congratulations! You have just trained your first models with Brisk.
HTML Report#
Whenever you train models with Brisk, all the results are saved in the results/
directory of your project (Brisk will create this directory if it doesn’t exist).
Here you should find a directory with a timestamp of when the experiments were run.
It should look like this:
tutorial/
└── results/
└── tutorial_results/
├── html_report/
│ └── index.html
│ └── ...
├── tutorial/
│ └── ...
└── config_log.md
config_log.md is a markdown file that contains the configuration used for the
experiments. It contains all of the information needed to reproduce the experiments
later. Even if you change the configuration files you can always reference this file
if you need to.
Each experiment group will have its own subdirectory, with the name of the group. There will be a subdirectory with analysis of the data split used for the group and a subdirectory for each experiment. This is where you can find the outputs of the methods you call in the workflow.
Finally there is an html_report directory. You can drag the index.html file
into your browser to view the report. Any of the evaluation methods provided by
Brisk you use in the workflow will be included in the report. This gives you a
quick overview of the performance of the models you trained. You can also look at
the distribution of features in the train test split.
If you want to learn more about Brisk, you can read the rest of the Getting Started with Brisk section. The User Guide section contains more detailed in depth information on specific features of Brisk.