Baseline model raise exception when using float32 value

[moshe-rl]

At 'Explain' mode, for regression task.
Other models complete training successfully.

errors.md file content:

Error for 1_Baseline Object of type 'float32' is not JSON serializable Traceback (most recent call last): File "/home/moshe/.local/lib/python3.6/site-packages/supervised/base_automl.py", line 970, in _fit trained = self.train_model(params) File "/home/moshe/.local/lib/python3.6/site-packages/supervised/base_automl.py", line 312, in train_model mf.save(model_path) File "/home/moshe/.local/lib/python3.6/site-packages/supervised/model_framework.py", line 395, in save fout.write(json.dumps(desc, indent=4)) File "/usr/lib/python3.6/json/init.py", line 238, in dumps **kw).encode(obj) File "/usr/lib/python3.6/json/encoder.py", line 201, in encode chunks = list(chunks) File "/usr/lib/python3.6/json/encoder.py", line 430, in _iterencode yield from _iterencode_dict(o, _current_indent_level) File "/usr/lib/python3.6/json/encoder.py", line 404, in _iterencode_dict yield from chunks File "/usr/lib/python3.6/json/encoder.py", line 437, in _iterencode o = _default(o) File "/usr/lib/python3.6/json/encoder.py", line 180, in default o.class.name) TypeError: Object of type 'float32' is not JSON serializable

[pplonski]

Thank you for reporting. Please provide minimal code example to reproduce.

[moshe-rl]

import numpy as np
import pandas as pd
from supervised.automl import AutoML
import tempfile
import os

# create random input data
data = np.random.randn(1000, 11)

# convert to dataframe
data_df = pd.DataFrame(data)
data_df.columns = [f'f{idx}' for idx in range(data.shape[1]-1)] + ['y']

# separate x and y
y_vec = data_df.pop('y').values
x_df = data_df

# train model
model = AutoML(
    results_path=os.path.join(tempfile.gettempdir(), 'aml_results__6'), 
    ml_task='regression',
    mode='Explain', 
    eval_metric='rmse',
    explain_level=2
)

##### HERE IS THE IMPORTANT PART #####
model.fit(x_df, y_vec.astype(np.float32))

Some things I've notice:

1. Some warnings raised:

Numerical issues were encountered when centering the data and might not be solved. Dataset may contain too large values. You may need to prescale your features.
Numerical issues were encountered when scaling the data and might not be solved. The standard deviation of the data is probably very close to 0. 
Numerical issues were encountered when centering the data and might not be solved. Dataset may contain too large values. You may need to prescale your features.

2. When using y_vec as is or even using y_vec.astype(np.float64) it works fine, the only problem is with float32
3. I digged a bit into the code, the problem seems to be with 'final_loss' key of the dict (desc) the code try to store as json

[pplonski]

@moshe-rl great job! Thank you!

[pplonski]

Should be fixed with @DanielR59 PR.

[csetzkorn]

just used latest version and encountered this issue as well ...

[pplonski]

Hi @csetzkorn,

could you provide minimal code to reproduce the issue?

[JacobMarley]

Yep. Getting this too. Problem caused by using xgb.