Read and create the booster model from a JSON file

[KhaitovR]

Hi guys!

I have looked through all the tasks related to saving models in JSON. Unfortunately, I could not figure out whether we can create a booster object from a JSON file?
I would really appreciate it if you could tell me how I can create a booster using json.

I think the size of the JSON model will be smaller than the STRING model.
Perhaps there is a lot of necessary information missing in the dump json model. Maybe that's why json weighs so much less.
I'll be glad if you correct me.

Reproducible example

import lightgbm as lgb
from sklearn.datasets import make_regression
from sys import getsizeof

X, y = make_regression(n_samples=10000, n_features=50, n_informative=20, noise=0.3, random_state=7)

model = lgb.train(
    params={
        'objective':'regression',
        'max_depth':8,
        'num_leaves':int(2**8*0.6),
        'verbose':-1
    },
    train_set=lgb.Dataset(X, y, free_raw_data=True)
)

print('lgb version:', lgb.__version__)
print('train shape (rows, columns)', (X.shape[0], X.shape[1]))
print('JSON-size mb:', format(getsizeof(model.dump_model(importance_type='gain'))/(1024*1024), '.4f'))
print('STRING-size mb:', format(getsizeof(model.model_to_string())/(1024*1024), '.4f'))

# lgb version: 3.3.2
# train shape (rows, columns) (10000, 50)
# JSON-size mb: 0.0006
# STRING-size mb: 0.7999

Additional Comments

Related task: #2604

[dishkakrauch]

@StrikerRUS could help us out with this issue please?

[jmoralez]

Hi. I see that the dump model functions were added in #97 as a way to analyze the underlying model. They're currently used in the trees to dataframe functions:

LightGBM/python-package/lightgbm/basic.py

Line 2820 in 44fe591

def trees_to_dataframe(self) -> pd_DataFrame:

LightGBM/R-package/R/lgb.model.dt.tree.R

Line 51 in 44fe591

lgb.model.dt.tree <- function(model, num_iteration = NULL) {

I don't think that format was meant to be an alternative to the original serialization but there's an ongoing discussion in #4887 about using JSON as the default (and in the future only) serialization format. But to answer your question, I don't think it's currently possible to restore a booster from JSON.

[KhaitovR]

I appreciate your quick response!
I hope that this feature will be added as soon as possible.
Should I close the task?

[jmoralez]

I think we can close it but feel free to comment on #4887 if you'd like to add anything to the proposal.

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.