I'm trying to utilize a custom scorer with the following code

def edge_score(y, y_pred):
    y_pred.name = 'y_pred'
    y.name = 'y'

    df = pd.concat([y_pred, y])

    df['sign_pred'] = df.y_pred.apply(np.sign)
    df['sign_true'] = df.y.apply(np.sign)
    df['is_correct'] = 0
    df.loc[
        df.sign_pred * df.sign_true > 0, 'is_correct'] = 1
    df['is_incorrect'] = 0
    df.loc[
        df.sign_pred * df.sign_true < 0, 'is_incorrect'] = 1
    df['is_predicted'] = df.is_correct   df.is_incorrect
    df['result'] = df.sign_pred * df.y
    df['edge'] = df.result.mean()
    output_errors = df[['edge']]
    output_errors.to_numpy()

    return np.average(output_errors)
edge = make_scorer(edge_score)

I get the following error

AttributeError: 'numpy.ndarray' object has no attribute 'name'

When I comment out the .name lines, I get the following error

TypeError: cannot concatenate object of type '<class 'numpy.ndarray'>'; only Series and DataFrame objs are valid

When I convert true and predictions to dataframe, I get the following error

y_pred = pd.DataFrame(y_pred)
y = pd.DataFrame(y)

AttributeError: 'DataFrame' object has no attribute 'y_pred'

https://scikit-learn.org/stable/modules/generated/sklearn.metrics.make_scorer.html#sklearn.metrics.make_scorer

CodePudding user response：

Change these lines of code

    df['sign_pred'] = df.y_pred.apply(np.sign)
    df['sign_true'] = df.y.apply(np.sign)

to these:

    df['sign_pred'] = np.sign(y_pred)
    df['sign_true'] = np.sign(y)

CodePudding user response：

You should first create a DataFrame with the two numpy arrays y and y_pred, and then perform all the operations.

def edge_score(y, y_pred):
    
    df = pd.DataFrame({"y":y,
                       "y_pred":y_pred})

    df['sign_pred'] = df.y_pred.apply(np.sign)
    df['sign_true'] = df.y.apply(np.sign)
    df['is_correct'] = 0
    df.loc[
        df.sign_pred * df.sign_true > 0, 'is_correct'] = 1
    df['is_incorrect'] = 0
    df.loc[
        df.sign_pred * df.sign_true < 0, 'is_incorrect'] = 1
    df['is_predicted'] = df.is_correct   df.is_incorrect
    df['result'] = df.sign_pred * df.y
    df['edge'] = df.result.mean()
    output_errors = df[['edge']]
    output_errors.to_numpy()

    return np.average(output_errors)

edge = make_scorer(edge_score)