config.rst

genno.config

Configuration

As shown in Concepts and usage <describe-tasks>, a .Computer can be populated programmatically. genno can also read a simple configuration format in which settings and tasks are specified.

Overview

Ways to configure

Configuration is read through the .Computer.configure method, or the top level genno.configure function.

Both methods accept either keyword arguments specifying the configuration values, or a path keyword argument that gives the path to a .json or .yaml-format file. For instance, the following are all equivalent:

# Create the computer before configuring from a JSON file
c = Computer()
c.configure(path="config-W.json")

# Create and configure from a YAML file in one step
c = Computer(path="config-W.yaml")

# Pass a data structure to configure()
info = dict(
  general=[
    dict(comp="product", key="W:a-b-c-d", inputs=["X::", "Y:d"], sums=True)
  ]
)
c = Computer()
c.configure(**info)

# Use the API to add a computation directly
c = Computer()
# add_product() infers the dimensions of W will be a-b-c-d
c.add_product("W", "X:a-b-c-d", "Y:d", sums=True)

…with the following file contents:

general:
- comp: product
  key: W:a-b-c-d
  inputs: ["X::", "Y:d"]
  sums: true

{
  "general": [
    {
      "comp": "product",
      "key": "W:a-b-c-d",
      "inputs": ["X::", "Y:d"],
      "sums": true
    }
  ]
}

Global- and specific configuration

Configuration is either global or specific to a certain Computer. For instance, config-units configuration is global; it affects all Computers, and can be set using either genno.configure or .Computer.configure. On the other hand, other configuration such as config-files adds tasks to a specific Computer, so it can only be set using .Computer.configure.

genno.configure

Custom handlers

The configuration file is divided into sections. Generally each section contains a list of items, and each item is itself a mapping; see the built-in sections listed below. .genno.config has one handler function for each section, and is extensible. For instance, the genno compatibility module for pyam <config-pyam> defines a handler for the section iamc:; or the separate package ixmp defines a handler for the sections filters: and rename_dims:.

The .handles decorator can be used to mark a custom function that handles a custom configuration section:

from genno.config import handles

@handles("my-section")
def custom_handler(c: Computer, info):
    print(f"Handle {info['name']}")
    print(f"  with inputs {repr(info['inputs'])}")

    # Use a default value for one setting
    key = info.get("key", "foo")
    print(f"Output key: {key}")

    # Manipulate the Computer instance `c` in some way
    c.add("… etc.")

# This section is handled by genno's built-in handlers
general:
- comp: product
   key: W:a-b-c-d
   inputs: ["X::", "Y:d"]
   sums: true

# These items are handled by the custom handler
my-section:
- name: item-a
  inputs: [X, Y]
- name: item-b
  key: bar
  inputs: [W, Z]

handles

HANDLERS

STORE

Specific sections

aggregate:

aggregate

Computer-specific configuration.

Invokes .Computer.aggregate add tasks with .computations.aggregate or .computations.sum, computing sums across labels within one dimension of a quantity. Each entry contains:

_quantities: list of 0 or more keys

Quantities to be aggregated. The full dimensionality of the key(s) is inferred.

_tag: (str)

New tag to append to the keys for the aggregated quantities.

_dim: (str)

Dimensions on which to aggregate.

Note the leading underscores. This is to distinguish these from all other keys, which are treated as group names. The corresponding values are lists of labels along the dimension to sum.

Example

aggregate:
- _quantities: [foo, bar]
  _tag: aggregated
  _dim: a

  baz123: [baz1, baz2, baz3]
  baz12: [baz1, baz2]

If the full dimensionality of the input quantities are foo:a-b and bar:a-b-c, then .add_aggregate creates the new quantities foo:a-b:aggregated and bar:a-b-c:aggregated. These new quantities have the new labels baz123 and baz12 along their a dimension, with sums of the indicated values.

alias:

alias

Computer-specific configuration.

This section simply makes the output of one task available under another key.

alias:
  "foo:x-y": "bar:x-y"
  "baz:x-y": "bar:x-y"

Caching

Computer-specific configuration that controls the behaviour of functions decorated with .Computer.cache.

cache_path: (pathlib.Path, optional)

Base path for cache files. If not provided, defaults to the current working directory.

cache_skip: (bool, optional)

If True, existing cache files are never used; files with the same cache key are overwritten.

combine:

combine

Computer-specific configuration.

Invokes .Computer.add_combination to add tasks with computations.combine, computing a weighted sum of multiple Quantities. Each item contains:

key:

Key for the new quantity, including dimensionality.

inputs: (list of dict)

Inputs to the weighted sum. Each dict contains:

quantity: (required)

Key for the input quantity. .add_combination infers the proper dimensionality from the dimensions of key plus dimension to select on.

select: (dict, optional)

Selectors to be applied to the input quantity. Keys are dimensions; values are either single labels, or lists of labels. In the latter case, the sum is taken across these values, so that the result has the same dimensionality as key.

weight: (int, optional)

Weight for the input quantity; default 1.

Example

For the following YAML:

combine:
- key: foo:a-b-c
  inputs:
  - quantity: bar
    weight: -1
  - quantity: baz::tag
    select: {d: [d1, d2, d3]}

….add_combination infers:

foo_abc =  − 1 × bar_abc + 1 × ∑_{d ∈ {d1, d2, d3}}baz_abcd^(tag) ∀ a, b, c

default: key/task

default

Computer-specific configuration.

This sets .Computer.default_key, used when .get is called without arguments.

files: input

files

Computer-specific configuration.

Invokes .Computer.add_file to add .computations.load_file. If the path: key is a relative path, it is resolved relative to the directory that contains the configuration file, else the current working directory.

files:
- path: ./input0.csv
  key: d_check
# 'dims' argument can be supplied as list or dict
- path: ./input1.csv
  key: input1-0
  dims: [i, j_dim]  # Omit extra dimension 'foo'
- path: ./input1.csv
  key: input1-1
  dims: {i: i, j_dim: j}

general:

general

Computer-specific configuration.

This is, as the name implies, the most generalized section. Each item contains:

comp:

Refers to the name of a computation that is available in the namespace of genno.computations, or custom computations registered by compatibility modules or third-party packages. See Computer.add and Computer.get_comp. E.g. if "product", then .Computer.add_product is called, which also automatically infers the correct dimensions for each input.

key:

The key for the computed quantity.

inputs:

A list of keys to which the computation is applied.

args: (dict, optional)

Keyword arguments to the computation.

add args: (dict, optional)

Keyword arguments to .Computer.add itself.

report:

report

Computer-specific configuration.

A ‘report’ is a concatenation of 1 or more other quantities.

Example

report:
- key: foo
  members: [X, Y]

units:

units

Global configuration.

Sub-keys:

replace: (mapping of str -> str)

Replace units before they are parsed by pint <pint:index>. Added to .REPLACE_UNITS.

define: (str)

Multi-line block of unit definitions, added to the pint application registry so that units are recognized. See the pint documentation on defining units <pint:defining>.

units:
  replace:
    dollar: USD
  # YAML multi-line string
  define: |-
    pp = [person]
    tiny = 0.1 millimetre