Model serialization#
There are two kinds of serialization, binary and json.
Caution
Serialized models should be loaded with the same version of the library that was used to create them. There are currently no guarantees that models will be compatible across versions.
Binary#
Saving#
reductionml.Workspace.serialize
from reductionml import Workspace
config = {"entryReduction": {"typename": "Coin"}, "globalConfig": {}}
workspace = Workspace.create_from_config(config)
model_data = workspace.serialize()
Binary models can be created at the end of a training run using the --output-model option:
-o, --output-model <OUTPUT_MODEL>
Loading#
reductionml.Workspace.create_from_model
from reductionml import Workspace
model_data = ... # load from file or other source
workspace = Workspace.create_from_model(config)
Binary models can be loading for a training run using the --input-model option:
-i, --input-model <INPUT_MODEL>
Json#
Caution
While a very powerful and helpful tool, extreme care must be taken if the json model is edited by hand. No guarantees are made that any changes will load back in correctly. If you do edit the model, you should understand all the implications of the changes you are making.
Saving#
reductionml.Workspace.serialize_to_json
from reductionml import Workspace
config = {"entryReduction": {"typename": "Coin"}, "globalConfig": {}}
workspace = Workspace.create_from_config(config)
model_data = workspace.serialize_to_json()
To create a json model file using the command line you must have an existing binary model and then it can be converted to json with:
reml export-model <INPUT_MODEL>
Loading#
reductionml.Workspace.create_from_json_model
from reductionml import Workspace
model_data = ... # load from file or other source
workspace = Workspace.create_from_json_model(config)
To convert a json model back into a binary model the following command can be used:
reml import-model <INPUT_FILE> --output-model <OUTPUT_MODEL>