polars-upgrade

Automatically upgrade your Polars code so it's compatible with future versions.

Installation

Easy:

pip install -U polars-upgrade

Usage (command-line)

Run

polars-upgrade my_project --target-version=0.20.31

from the command line. Replace 0.20.31 and my_project with your Polars version, and the name of your directory.

NOTE: this tool will modify your code! You're advised to stage your files before running it.

Usage (pre-commit hook)

-   repo: https://github.com/MarcoGorelli/polars-upgrade
    rev: 0.3.6  # polars-upgrade version goes here
    hooks:
    -   id: polars-upgrade
        args: [--target-version=0.20.31]  # Polars version goes here

Usage (Jupyter Notebooks)

Install nbqa and then run

nbqa polars_upgrade my_project --target-version=0.20.31

Usage (library)

In a Python script:

from polars_upgrade import rewrite, Settings

src = """\
import polars as pl
df.select(pl.count())
"""
settings = Settings(target_version=(0, 20, 4))
output = rewrite(src, settings=settings)
print(output)

Output:

import polars as pl
df.select(pl.len())

If your snippet does not include import polars or import as pl, then you will also need to provide pl and/or polars to aliases, else polars-upgrade will not perform the rewrite. Example:

from polars_upgrade import rewrite, Settings

src = """\
df.select(pl.count())
"""
settings = Settings(target_version=(0, 20, 4))
output = rewrite(src, settings=settings, aliases={'pl'})
print(output)

Output:

df.select(pl.len())

Supported rewrites

Version 0.18.12+

- pl.avg
+ pl.mean

Version 0.19.0+

- df.groupby_dynamic
+ df.group_by_dynamic
- df.groupby_rolling
+ df.rolling
- df.rolling('ts', period='3d').apply
+ df.rolling('ts', period='3d').map_groups
- pl.col('a').rolling_apply
+ pl.col('a').rolling_map
- pl.col('a').apply
+ pl.col('a').map_elements
- pl.col('a').map
+ pl.col('a').map_batches
- pl.map
+ pl.map_batches
- pl.apply
+ pl.map_groups
- pl.col('a').any(drop_nulls=True)
+ pl.col('a').any(ignore_nulls=True)
- pl.col('a').all(drop_nulls=True)
+ pl.col('a').all(ignore_nulls=True)
- pl.col('a').value_counts(multithreaded=True)
+ pl.col('a').value_counts(parallel=True)

Version 0.19.2+

- pl.col('a').is_not
+ pl.col('a').not_

Version 0.19.3+

- pl.enable_string_cache(True)
+ pl.enable_string_cache()
- pl.enable_string_cache(False)
+ pl.disable_string_cache()
- pl.col('a').list.count_match
+ pl.col('a').list.count_matches
- pl.col('a').is_last
+ pl.col('a').is_last_distinct
- pl.col('a').is_first
+ pl.col('a').is_first_distinct
- pl.col('a').str.strip
+ pl.col('a').str.strip_chars
- pl.col('a').str.lstrip
+ pl.col('a').str.strip_chars_start
- pl.col('a').str.rstrip
+ pl.col('a').str.strip_chars_end
- pl.col('a').str.count_match
+ pl.col('a').str.count_matches
- pl.col("dt").dt.offset_by("1mo_saturating")
+ pl.col("dt").dt.offset_by("1mo")

Version 0.19.4+

- df.group_by_dynamic('ts', every='3d', truncate=True)
+ df.group_by_dynamic('ts', every='3d', label='left')
- df.group_by_dynamic('ts', every='3d', truncate=False)
+ df.group_by_dynamic('ts', every='3d', label='datapoint')

Version 0.19.8+

- pl.col('a').list.lengths
+ pl.col('a').list.len
- pl.col('a').str.lengths
+ pl.col('a').str.len_bytes
- pl.col('a').str.n_chars
+ pl.col('a').str.len_chars

Version 0.19.11+

- pl.col('a').shift(periods=4)
+ pl.col('a').shift(n=4)
- pl.col('a').shift_and_fill(periods=4)
+ pl.col('a').shift_and_fill(n=4)
- pl.col('a').list.shift(periods=4)
+ pl.col('a').list.shift(n=4)
- pl.col('a').map_dict(remapping={1: 2})
+ pl.col('a').map_dict(mapping={1: 2})

Version 0.19.12+

- pl.col('a').keep_name
+ pl.col('a').name.keep
- pl.col('a').suffix
+ pl.col('a').name.suffix
- pl.col('a').prefix
+ pl.col('a').name.prefix
- pl.col('a').map_alias
+ pl.col('a').name.map
- pl.col('a').str.ljust
+ pl.col('a').str.pad_end
- pl.col('a').str.rjust
+ pl.col('a').str.pad_start
- pl.col('a').zfill(alignment=3)
+ pl.col('a').zfill(length=3)
- pl.col('a').ljust(width=3)
+ pl.col('a').ljust(length=3)
- pl.col('a').rjust(width=3)
+ pl.col('a').rjust(length=3)

Version 0.19.13+

- pl.col('a').dt.milliseconds
+ pl.col('a').dt.total_milliseconds
- pl.col('a').dt.microseconds
+ pl.col('a').dt.total_microseconds
- pl.col('a').dt.nanoseconds
+ pl.col('a').dt.total_nanoseconds

(and so on for other units)

Version 0.19.14+

- pl.col('a').list.take
+ pl.col('a').list.gather
- pl.col('a').cumcount
+ pl.col('a').cum_count
- pl.col('a').cummax
+ pl.col('a').cum_max
- pl.col('a').cummin
+ pl.col('a').cum_min
- pl.col('a').cumprod
+ pl.col('a').cum_prod
- pl.col('a').cumsum
+ pl.col('a').cum_sum
- pl.col('a').cumcount
+ pl.col('a').cum_count
- pl.col('a').take
+ pl.col('a').gather
- pl.col('a').take_every
+ pl.col('a').gather_every
- pl.cumsum
+ pl.cum_sum
- pl.cumfold
+ pl.cum_fold
- pl.cumreduce
+ pl.cum_reduce
- pl.cumsum_horizontal
+ pl.cum_sum_horizontal
- pl.col('a').list.take(index=[1, 2])
+ pl.col('a').list.take(indices=[1, 2])
- pl.col('a').str.parse_int(radix=1)
+ pl.col('a').str.parse_int(base=1)

Version 0.19.15+

- pl.col('a').str.json_extract
+ pl.col('a').str.json_decode

Version 0.19.16+

- pl.col('a').map_dict({'a': 'b'})
+ pl.col('a').replace({'a': 'b'}, default=None)
- pl.col('a').map_dict({'a': 'b'}, default='c')
+ pl.col('a').replace({'a': 'b'}, default='c')

Version 0.20.0+

- df.write_database(table_name='foo', if_exists="append")
+ df.write_database(table_name='foo', if_table_exists="append")

Version 0.20.4+

- pl.col('a').where
+ pl.col('a').filter
- pl.count()
+ pl.len()
- df.with_row_count('row_number')
+ df.with_row_index('row_number')
- pl.scan_ndjson(source, row_count_name='foo', row_count_offset=3)
+ pl.scan_ndjson(source, row_index_name='foo', row_index_offset=3)
[...and similarly for `read_csv`, `read_csv_batched`, `scan_csv`, `read_ipc`, `read_ipc_stream`, `scan_ipc`, `read_parquet`, `scan_parquet`]

Version 0.20.5+

- df.pivot(index=index, values=values, columns=columns, aggregate_function='count')
+ df.pivot(index=index, values=values, columns=columns, aggregate_function='len')

Version 0.20.6+

- pl.read_excel(source, xlsx2csv_options=options, read_csv_options=read_options)
+ pl.read_excel(source, engine_options=options, read_options=read_options)

Version 0.20.7+

- pl.threadpool_size
+ pl.thread_pool_size

Version 0.20.8+

- df.pivot(a, b, c)
+ df.pivot(values=a, index=b, columns=c)

Version 0.20.11+

- pl.col('a').meta.write_json
+ pl.col('a').meta.serialize

Version 0.20.14+

- df.group_by_dynamic('time', every='2d', by='symbol')
+ df.group_by_dynamic('time', every='2d', group_by='symbol')
- df.rolling('time', period='2d', by='symbol')
+ df.rolling('time', period='2d', group_by='symbol')
- df.upsample('time', every='2d', by='symbol')
+ df.upsample('time', every='2d', group_by='symbol')

Version 0.20.17+

- pl.from_repr(tbl=data)
+ pl.from_repr(data=data)

Version 0.20.24+

- pl.col('a').rolling_min('2d', by='time')
+ pl.col('a').rolling_min_by(window_size='2d', by='time')
- pl.col('a').rolling_max('2d', by='time')
+ pl.col('a').rolling_max_by(window_size='2d', by='time')
- pl.col('a').rolling_mean('2d', by='time')
+ pl.col('a').rolling_mean_by(window_size='2d', by='time')
- pl.col('a').rolling_std('2d', by='time')
+ pl.col('a').rolling_std_by(window_size='2d', by='time')
- pl.col('a').rolling_var('2d', by='time')
+ pl.col('a').rolling_var_by(window_size='2d', by='time')
- pl.col('a').rolling_prod('2d', by='time')
+ pl.col('a').rolling_prod_by(window_size='2d', by='time')
- pl.col('a').rolling_sum('2d', by='time')
+ pl.col('a').rolling_sum_by(window_size='2d', by='time')

Version 0.20.29+

- df.join(df_right, how='outer')
+ df.join(df_right, how='full')
- df.join(df_right, how='outer_coalesce')
+ df.join(df_right, how='full', coalesce=True)

Version 0.20.31+

- pl.read_csv(file, dtypes=schema)
+ pl.read_csv(file, schema=schema)
- pl.SQLContext(eager_execution=True)
+ pl.SQLContext(eager=True)
- pl.col('a').top_k(k=2, maintain_order=True)
+ pl.col('a').top_k(k=2)

Notes

This work is derivative of pyupgrade - many parts have been lifted verbatim. As required, I've included pyupgrade's license.