Search code examples
pythonpandastime-seriesseabornregplot

regplot using datetime64 as the x axis

I have a dataframe looks like this:

date         score  
2017-06-04    90
2017-06-03    80
2017-06-02    70

When I tried this:

sns.regplot(x=date, y=score, data=df)

I got an error:

TypeError: reduction operation 'mean' not allowed for this dtype

The dtype for date is datetime64[ns], and int64 for the score column.

How can I covert the date column so that regplot will work?

Solution

  • Seaborn doesn't support datetimes in regplot but here's an ugly hack:

    df = df.sort_values('date')
df['date_f'] = pd.factorize(df['date'])[0] + 1
mapping = dict(zip(df['date_f'], df['date'].dt.date))

ax = sns.regplot('date_f', 'score', data=df)
labels = pd.Series(ax.get_xticks()).map(mapping).fillna('')
ax.set_xticklabels(labels)

    produces

    enter image description here

    This is the main approach used in time-series regression. If you have daily data, you code day 1 as 1 and increase the number as the days go by. This assumes you have a regularly-spaced time series.