Circular reference error

When trying to log pandas dataframe constructed via function, an error occurs. Using .copy(deep=True) doesn’t help.

Example code:

from sklego.model_selection import TimeGapSplit
from datetime import timedelta

cv = TimeGapSplit(
    date_serie=df.reset_index("time").loc[:, "time"],
    train_duration=timedelta(days=7),
    valid_duration=timedelta(days=1),
    gap_duration=timedelta(days=0),
    n_splits=3,
)


table_plot = cv.summary(df)
title = "cross-validation folds"
series = "1"
Logger.current_logger().report_table(title=title, series=series, table_plot=table_plot)

Actual result:

ValueError: Circular reference detected

UPD: This did not help:

df_ = cv.summary(df_dev)
_df = pd.DataFrame(data=df_.values, index=df_.index, columns=df_.columns)

table_plot = _df
title = "cross-validation folds"
series = "1"
Logger.current_logger().report_table(title=title, series=series, table_plot=table_plot)

[ColdTeapot273K]

@jkhenning tested on 1.1.5rc4, works, nice job!

Wouldn’t call it too readable, but that’s on pandas/plotly team. At least it doesn’t break now so doesn’t make you write ad hoc boilerplate for type conversion.

[jkhenning]

@ColdTeapot273K we might have found a way to support it by using to_json() and letting pandas handle it 🙂 - I’ll update when an RC with this fix is available