DataClass remove inf vals

Question

I have a data-class

from dataclasses import dataclass

@dataclass
class ModelResult:
    scorecard_calcs: dict

This one is then used in the Model class

import pandas as pd
import numpy as np

class Model:
    def predict(self, input):
        scorecard_calcs = pd.DataFrame([{'val':np.inf}, {'val':-np.inf}]).to_dict()
        return ModelResult(scorecard_calcs=scorecard_calcs)

Somewhere else this will be converted into json, but inf is not valid json.

How can I change the ModelResult dataclass, so it automatically replaces all inf values with 1.7976931348623157e+308 which is valid json.

Answer 1

Here is how I did it with a __post_init__ as suggested by @balderman

from dataclasses import dataclass
import numpy as np

ALMOST_INF = 1.7976931348623157e308

@dataclass
class ModelResult:
    scorecard_calcs: dict
    pd: float

    @staticmethod
    def replace_inf(val):
        if not isinstance(val, (int, float, complex)):
            return val

        if np.isposinf(val):
            return ALMOST_INF

        if np.isneginf(val):
            return -ALMOST_INF

        return val

    def __post_init__(self):
        self.pd = self.replace_inf(self.pd)

        scorecard_calcs = pd.DataFrame(self.scorecard_calcs)
        scorecard_calcs = scorecard_calcs.applymap(self.replace_inf)
        self.scorecard_calcs = scorecard_calcs.to_dict()

This works, since I know for a fact that scorecard_calcs is a dict which was previously a pandas data-frame.