348. Jinja2: Model Configuration Version Control

Creating Templates

Jinja2 is a template engine for Python which can be useful when you want to keep track of model training configuration versions.

Here is one way to Implement it by combining them with a YAML file.

1. Create Jinja2 Template: template.yaml.j2

DATASET:
  TEST_SET: '{{TEST_SET_DIR | default('./test')}}'
  TRAIN_SET: '{{TRAIN_SET_DIR | default('./train')}}'
MODEL:
  NAME: FCN
TRAIN:
  BASE_SIZE: {{BATCH_SIZE | default(4)}}
  EPOCHS: {{EPOCHS | default(25)}}
TEST:
  BASE_SIZE: {{BATCH_SIZE | default(4)}}
  EPOCHS: {{EPOCHS | default(25)}}

2. Render Configuration YAML file from template

from jinja2 import Template

#File name to export rendered file
render_to  ='configured.yaml'

# Open jinja2 template which we've just created
with open('model_config.yaml.j2') as file_:
    template = Template(file_.read())

# Set hyper-parameter configuration:
# It might be useful for the user to specify these through a GUI
param_config = {
    "TEST_SET": "./test_set",
    "TRAIN_SET": "./train_set",
    "BASE_SIZE": "16",
    "EPOCHS": "25",
}

#Render using template
rendered_conf = template.render(param_config)

#Write file
with open(render_to, 'w') as f:
    f.write(rendered_conf)    

Generated Yaml FILE

DATASET:
  TEST_SET: './test'
  TRAIN_SET: './train'
MODEL:
  NAME: FCN
TRAIN:
  BASE_SIZE: 4
  EPOCHS: 25
TEST:
  BASE_SIZE: 4
  EPOCHS: 25

Now you can use the generated YAML file to start the training.