Parameters

To parameterize a step, you need parameters. They can be float, integer, string or boolean type.

With valohai-utils, you define the parameters in the call to the prepare method. Feed the default_parameters argument with a key/value dictionary of the parameters.

  • key is the name of the parameter. Used in the code, the YAML and the Valohai UI.

  • value defines the default value of the parameter and also it’s type.

Note

Empty values are not supported as they have no type.

Defining parameters with valohai-utils solves two problems:

  • Parsing the command-line overrides

  • Managing the duplicate definitions between Python & YAML

train.py

Here we define a step train, with a parameter learning-rate.

import valohai


params = {
    "learning-rate": 0.001,
}

valohai.prepare(
    step="train",
    default_parameters=params,
)

This key/value pair…

params = {
    "learning-rate": 0.001
}

…will be transpiled into the following YAML

- name: learning-rate
  default: 0.001
  optional: false
  type: float

Accessing values

Once you have defined a parameter using the prepare method, you can access it in your code by referring to the parameter name.

lr = valohai.parameters("learning-rate").value

Overriding values

All parameters defined by the prepare method always have a default value.

There are two ways to override the default value:

  • Command-line parameter (local)

  • Valohai UI or CLI (remote)

Example (local):

python train.py --learning-rate=.002

Example (remote):

vh yaml step train.py
vh exec run -a train --learning-rate=.002