Layout

Layout is a mechanism in MLComp to customize your reports.

MLComp already includes some layouts.

You can find them at http://localhost:4201/reports/layouts

The structure is the following:

extend: base # an optional field

# To choose the best epoch
metric:
  name: str
  minimize: True/False

items: list of metrics to log during training
layout: list of layout components

extend

Name of an existing layout. Items and Layouts will be merged

items

Each item is a dictionary with a required key type.

By key each item is available in layout components ( key field )

Possible types:

• series

  • key: str [required]

    source of a metric

layout

Customizes appearance in the UI.

type is required to choose a component

Possible components:

• panel

  • title: str [required]

  • expanded: True/False

  • parent_cols: int

  • cols: int

  • row_height: int

  • items: list of other components

  • table: True/False

    Use a table structure instead of Grid List. In this case child items are not put on new rows, but child components get more vertical space.

• blank

  • cols

  • rows

• series

  • source: str [required]

  • multi: True/False

    Either plot many series on a single plot or divide them into individual plots.

  • group: train/valid

    if you want to see only specific group

  • rows: int

  • cols: int

• table

  • source: list[str]

    list of metrics to compare

  • rows: int

  • cols: int

• img

  • source: str [required]

  • rows: int

  • cols: int