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:
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 |
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")