Checks <input type="c

Oh right. .cum_count() start value changed from <code

ewm_mean produces different results in 0.20.21 than 0.19.14 about polars HOT 5 CLOSED

IanAtDazed commented on May 24, 2024

ewm_mean produces different results in 0.20.21 than 0.19.14

from polars.

Comments (5)

MarcoGorelli commented on May 24, 2024 1

Thanks for reporting, could you make a more minimal example that only uses ewm_mean please? I suspect it's another part of the computation which changed. I think cum_count changed at some point

from polars.

cmdlineluser commented on May 24, 2024 1

Oh right. .cum_count() start value changed from 0 to 1 in 0.20.4 ... as @MarcoGorelli suggested.

So the code would have to be changed from < 9 to <= 9 or < 10

Although if cum_count null behaviour also changed, something else may need adjusting to handle that?

from polars.

IanAtDazed commented on May 24, 2024 1

Thank you! My problem is solved! :-)

from polars.

cmdlineluser commented on May 24, 2024

Things appear to have changed in 0.20.4 and it is before .ewm_mean() is called:

import polars as pl

values = [
    143.15,143.1,143.06,143.01,143.03,143.09,143.14,143.18,
    143.2,143.2,143.2,143.31,143.38,143.35,143.34,143.25,
    143.33,143.3,143.33,143.36
]

(
    pl.DataFrame({'value': values})
      .with_columns(
          pl.when(pl.col('value').cum_count() < 9)
            .then(pl.col('value').head(9).mean())
            .otherwise(pl.col('value'))
      )
      .to_dict(as_series=False)
)

0.20.3

{'value': [143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.2,
  143.2,
  143.31,
  143.38,
  143.35,
  143.34,
  143.25,
  143.33,
  143.3,
  143.33,
  143.36]}

0.20.4

{'value': [143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.10666666666668,
  143.2,              # this was 143.10666666666668 in 0.20.3
  143.2,
  143.2,
  143.31,
  143.38,
  143.35,
  143.34,
  143.25,
  143.33,
  143.3,
  143.33,
  143.36]}