Cannot serialize numpy.float32

Hi,

I have an issue serialising a model that was fit with a column of type float32.

Version: fbprophet == 0.7.1 pystan == 2.19.1.1

Example code:

import pandas as pd
import numpy as np
from fbprophet import Prophet
from fbprophet.serialize import model_to_json

df = pd.DataFrame()
df["ds"] = pd.date_range(start="2020-01-01", end="2021-01-01", freq="1MS")
df["y"] = np.random.rand(df.shape[0])
df["y"] = df["y"].astype("float32")

model = Prophet()
model.fit(df)
model_json = model_to_json(model)

I did a bit of investigation:

# after returning the model_json before serialization:
import json
for key in model_json:
    try:
        json.dumps(model_json[key])
    except TypeError:
        print(key)

This returns only y_scale and type(model_json["y_scale"]) => numpy.float32.

Within serialize.py y_scale is in the list SIMPLE_ATTRIBUTES, as such it is serialised directly rather than using np.to_list()

This may relate to pull request #1656 and issue #1701

Thanks!

[stanton119]

Yes! that’s the problem. I’m on Pandas 1.2.0 and I’m getting the .max() as a numpy type:

y_scale = df["y"].max()
print(type(y_scale))  # <class 'numpy.float32'>

I think your commit will solve it.

[bletham]

This fix has been pushed to pypi in the new v1.0 release.