The Rust crate polars
has its own changelog.
-
bug fix
- fix bug in cum_sum #604
-
feature
- DataFrame.describe method #606
- Multi-level sorting of a DataFrame #607
- Expand functionality of Expr.is_in #614
- Csv-parser low_memory option #615
- Allow expressions in
pl.arange
#611
-
performance
- sort().reverse() optimization #605
- bug fix
- null handling in mean, std, var, and cov aggregations. #595
- rev-mapping of categorical stored duplicates. #595
- fix memory surge after csv-parsing #593
-
bug fix
- Throw error on join from different string cache #584
- fix covariance of array with null values #585
-
feature
- Series describe method #569
- dsl: take, arg_unique, unique
- allow lazy expressions in Eager API # 588
- describe Series
-
performance
- fix accidental expensive appends #592
- remove chunk_id from ChunkedArray #593
-
bug fix
- ensure column name persist after pyarrow cast #563
- make sure that
agg_list
maintains dtype #567 - fix panic in physical dispatch of Date dtypes
-
feature
- Implicitly Cast dtypes to temporal types in csv parser #560
- Series describe method #569
-
performance
- Cache and improve window functions performance #570
-
bug fix
- fix bug with pyarrow chunkedarray: #545
-
feature
- DataFrame.apply method
- Make a Series a Literal
- Make None a Literal
-
performance
- Update arrow
- faster iterators
- faster kernels
- Update arrow
-
bug fix
- fix bug in downsample: #537
-
feature
- cast categorical in csv parser: #533
- add many groupby-context aware operations: #534
- dowcast by month: #537
-
performance
- improve iterator in no null case: #538
- remove indirection: #536
-
bug fix
- fix bug in vectorized hashing algorithm that affected groupbys with null values: #523
- fix bug in downsample: 528
- change median algorithm: #527
-
feature
- use lazy groupby API/DSL in eager API: #522
- make sort groupby-context aware: #522
-
performance
- improve sort algorithms for sort and argsort: #526
-
performance
- [python | rust] multi-threaded outer join
- [python | rust] better performance in groupby on multiple keys (faster hashmap comparisons)
- [python | rust] better performance in multi column joins
-
bug fix
- [python] make horizontal aggregations null aware
-
feature
- [python | rust] Downsample by week
- [python | rust] join by unlimited columns
- [python]
Create a list Series directly. - [python] Create DataFrame from np.ndarray
-
bug fix
- [python] pandas to polars date64, maintain time information
- [python] fix bug in Date64 Series.year
- [python] fix bug Series.mean (did not correct for null values) #484
- [python | rust ] fix bug in rolling windows #484
- [python | rust ] fix bug lazy csv parser #459
-
feature
- [python | rust] Series methods
- Series.week
- Series.weekday
- Series.arg_min
- Series.arg_max
- Series.shape
- [python | rust] Series methods
-
bug fix
- [python] More pyarrow -> polars conversions.
-
feature
- [python] DataFrame methods: [ shift_and_fill].
- [python] eager: sum, min, max, mean horizontal aggregation.
- performance
- [python | rust] arrow arrays have a layer of indirection less; 10/20% performance improvement
-
name change: Python bindings module renamed from pypolars to polars
-
name change: Python bindings package renamed from py-polars to polars
-
feature
- [python] lazy: DataFrame methods: [ tail, first, last ].
- [python] eager: DataFrame fold for horizontal aggregation.
- [python] eager: Series methods: [median, quantile, is_in, to_frame]
- [python] eager: iterate over groupby and yield groups' DataFrames
- [python] eager: groupby.get_group('value')
- [python] add parquet compression
- [python] shift_and_fill expression
- [python] implicitly download raw files from the web in
read_parquet
,read_csv
. - [python | rust] methods for local peak finding in numerical series
- [python | rust] faster query optimization due to local memory arena's.
- [rust] reduce default compile time by making less features default.
- [python | rust] Series zip_with implicitly cast to supertype.
- [python | rust] window functions have a
min_periods
argument to control when to compute a result
-
bug fix
- [python] support file buffers for reading and writing csv and parquet
- [python | rust] fix csv-parser: allow new-line character in a string field
- [python | rust] don't let predicate-pushdown pass shift | sort operation to maintain correctness.
- performance
- [python | rust] use mimalloc global allocator
- [python | rust] undo performance regression on large number of threads
- bug fix
- [python | rust] fix accidental over-allocation in csv-parser
- [python] support agg (dictionary aggregation) for downsample
- performance
- [python | rust] categorical type groupby keys (use size hint)
- [python | rust] remove indirection layer in vector hasher
- [python | rust] improve performance of null array creation
- bug fix
- [python] implement set_with_mask for Boolean type
- [python | rust] don't panic (instead return null) in dataframe aggregation
std
andvar
- other
- [rust] internal refactors
- bug fix
- [python] fix various pyarrow related bugs
- feature
- [python] render html tables
- performance
- [python] default to pyarrow for parquet reading
- [python | rust] use u32 instead of usize in groupby and join to increase cache coherence and reduce memory pressure.