HypML: Hypothesis Markup Language

HypML is an open standard for specifying a machine learning task (a hypothesis). Using a .hypml document, you can separate the tasks of algorithmic design and implementation.

You can read more about the project here.

Basic Structure

HypML has three required sections:

• task
• target
• features

Here’s a simple example:

task: classification

target:
  Positive class label: 1
  Negative class label: 0

features:
  Feature 1:
    type: float
  Feature 2:
    type: bool
  Feature 3:
    type: float

You can also extend your HypML document with these optional sections:

• meta
• metrics

Comments

# Comments are cool too!

You can indicate a line in your HypML document is a comment by starting it with the # character. You can use comments to include additional information for anyone reading your HypML document, however these lines will be ignored during processing.

Sections

task

This section specifies details of the machine learning task to be implemented.

Classification

task: classification

Specifies your hypothesis as a classification task. This means your target will need to include at least two class labels.

Regression

task: regression

Specifies your hypothesis as a regression task. This means your target will need be a single continuous number.

target

TODO

features

features:
  Name of Feature 1:
    type: Type of Feature 1
  Name of Feature 2:
    type: Type of Feature 2

Here you can represent each of your features by name and indicate the type of the feature.

You must specify at least 1 feature, and you must specify the type of each.

Available types:

• integer 1, -2, 300, 42131, -10
• float 1.001, -2.23, 400123.2393, -0.0001
• bool 1, 0, True, False
• string a, b, c, d (assumed as categorical text)
• text aaa aaa aaa, bb bbbb bbbbb, c ccc cc (assumed free-text)

meta

meta:
  title: Idea v1
  version: 1
  author: Blair Hudson
  date: 20181105
  description: My awesome hypothesis 

If you include the meta section, you must include at least a title.

Available meta properties:

• title Short hypothesis name
• version 1, 2, 3.0, 3.2 (any integer or float)
• author Your name
• date 20180101 (YYYYMMDD)
• description Longer format description of what your hypothesis is about

metrics

metrics:
  metric 1:
    min: 0.9
  metric 2:
    max: 0.2
  metric 3:
    min: 0.4
    max: 0.6
  metric 4

You can include any number of metrics, specifying a minimum, maximum or both minimum and maximum values to be expected, or neither if the metric is simply for information.

Available metrics:

• accuracy
• auc