Thanks for the report

@amueller Shall I take this ?

I can take this up. NYC WiMLDS

I'm going to take a stab at this and will edit this comment for updates. -NYC WiMLDS

Update #1: I'm currently scouring through both calibration.py and isotonic.py files, and I can't figure out what makes the code trigger for infinite values. For example, y_min and y_max are default set to None and the if block on isotonic.py handles for when y_min/y_max is None, set them to infinite values. I am looking through the predict_proba code for predicting probabilities, and this might be the piece of code that sets it off. I'll do more testing on predict_proba() to see if any infinite values come up. @jay-z007, any thoughts from your end?

`def predict_proba(self, X):
"""Posterior probabilities of classification

    This function returns posterior probabilities of classification
    according to each class on an array of test vectors X.

    Parameters
    ----------
    X : array-like, shape (n_samples, n_features)
        The samples.

    Returns
    -------
    C : array, shape (n_samples, n_classes)
        The predicted probas. Can be exact zeros.
    """
    n_classes = len(self.classes_)
    proba = np.zeros((X.shape[0], n_classes))

    df, idx_pos_class = self._preproc(X)

    for k, this_df, calibrator in \
            zip(idx_pos_class, df.T, self.calibrators_):
        if n_classes == 2:
            k += 1
        proba[:, k] = calibrator.predict(this_df)

    # Normalize the probabilities
    if n_classes == 2:
        proba[:, 0] = 1. - proba[:, 1]
    else:
        proba /= np.sum(proba, axis=1)[:, np.newaxis]

    # XXX : for some reason all probas can be 0
    proba[np.isnan(proba)] = 1. / n_classes

    # Deal with cases where the predicted probability minimally exceeds 1.0
    proba[(1.0 < proba) & (proba <= 1.0 + 1e-5)] = 1.0

    return proba`

@jay-z007 @jfrank94 please make sure you coordinate.

I will take this one

Me and @ImenRajhi, so far found out that it is caused by scipy.interpolate.interp1d() function that is used in isotonic.py in line249:

self.f_ = interpolate.interp1d(X, y, kind='linear', 
                                           bounds_error=bounds_error) 

It is caused by small values of X input. If you add fill_value="extrapolate" to that line, you would get Error: .../scipy/interpolate/interpolate.py:609: RuntimeWarning: overflow encountered in true_divide

  slope = (y_hi - y_lo) / (x_hi - x_lo)[:, None] 

Since, (x_hi - x_lo) is super small, and dividing by it, causing overflow.

@NurzatRakhman what versions of NumPy scipy and scikit-learn are you running?

@thomasjpfan

Libraries: [this is the latest]
scipy==1.4.1
scikit-learn 0.23.dev0
numpy 1.16.5

When you run code with given data, set seed to split method to 22
sss = StratifiedShuffleSplit(n_splits = 10, test_size = 0.2, random_state=22)

Then you will see that predicted probabilities for index = [1111, 2057] return "inf" result and you can observe the behaviour above that I described.

I'm working on this issue.

Take