Skip to content

Override Defaults via Config File

To utilise a YAML config file instead of CLI overriding, you can following the examples in this section. And, as always, to understand the defaults and what can be overridden, use canari_ml train --help.

Examples

Below is an example configuration:

configs/train/custom_train.yaml
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
# @package _global_

defaults:                                   # (1)!
  - ../preprocess/train_demo_dataset.yaml   # (2)!
  - /train                                  # (3)!
  - _self_                                  # (4)!

train:
  dataset: preprocessed_data/train_demo_dataset/03_cache_demo_dataset/cached.DAY.north.json
  name: demo_train
  seed: 42
  epochs: 2
  workers: 4
  batch_size: 4
  shuffling: true
  wandb_group: demo_unet
  wandb_project: CANARI_Training

model:
  model_name: unet
  network:
    filter_size: 3
    n_filters_factor: 0.1
  litmodule:
    criterion:
      loss_type: mse
  1. Always define defaults in the header of your custom config file.
  2. Define path to the config file used for training.
  3. Define /train to inherit from the default configuration.
  4. Override the above defaults with values from this file. (The order matters, _self_ should be defined last to override previous configs in this list).

You can run training with this config file using:

canari_ml train -cd configs/train/ -cn custom_train.yaml