Info mode

This notebook documents the capabilities of the MML info mode.

# giving no tasks will deliver no output information
# side note: this is the default configuration when entering solely 'mml' allowing to test a successful installation
!mml info tasks=none

For further explanation we need some installed tasks, you may download and store them automatically by yourselves when running mml create tasks=fake. See the create mode tutorial for more details on this feature. Assume these tasks have been installed:

# giving some tasks will automatically present task information and create a sample grid
!mml info tasks=fake

The output provides us with an overview on the most important features of the tasks we used. Further it generated a sample grid with one image per task. Let’s inspect that image:

# the grid was placed here: /home/scholzpa/Documents/exp/mml_results/default/PLOTS/sample_grid/grid_0079.png. as can be seen by the log
from IPython import display

display.Image("/home/scholzpa/Documents/exp/mml_results/default/PLOTS/sample_grid/grid_0079.png")

Be aware that the default preprocessing configuration has been applied!

Info mode further support inspecting

• trained models,

• study results of a (persistent!) hyperparameter optimization

Use the respective mode.subroutines when calling mml and either specify the study or reuse some models

!mml info mode.subroutines=[hpo] hpo=sql mode.study_name=pami2_fed_hpo_03

!mml info mode.subroutines=[models] reuse.models=test tasks=fake