Renate’s output

The result of a Renate update job is written to the folder specified via the next_state_url attribute. This folder contains the Renate state. It will be required for the next update job.

The Renate state folder contains three files:

Renate Dataset Overview

File	Description
model.ckpt	This is the checkpoint of the trained model and the only file required for deployment. Use and load this file to make predictions.
learner.ckpt	This contains the state of the Renate updater. Only used by Renate.
hpo.csv	A summary of all previous updates. The update_id with highest value refers to the last update step. Among other things, it contains information about selected hyperparameters and logged metrics. It might be used in the next update step to accelerate the hyperparameter tuning step.

Loading the Updated Model

In the following, we refer with model_fn to the function defined by the user in the Renate config file.

Output Saved Locally

If output_state_url is a path to a local folder, loading the updated model can be done as follows:

from renate.defaults import input_state_folder, model_file

my_model = model_fn(model_file(input_state_folder(output_state_url)))

Output Saved on S3

If the Renate output was saved on S3, the model checkpoint model.ckpt can be downloaded from

from renate.defaults import input_state_folder, model_file

print(model_file(input_state_folder(output_state_url)))

and then loaded via

my_model = model_fn("model.ckpt")