Releases: Safe-DS/Library
Releases · Safe-DS/Library
v0.26.0
0.26.0 (2024-05-29)
Features
Table.count_row_if
(#788) (4137131), closes #786- added method to load pretrained models from huggingface (#790) (dd8394b)
- infer input size of forward and LSTM layers (#808) (098a07f)
- outline around dots of scatterplot (#785) (ee8acf7)
- remove output conversions (#792) (46f2f5d), closes #732
- shorten some excessively long names (#787) (1c3ea59), closes #772
- specify column names in constructor of table transformers (#795) (69a780c)
- store window size and forecast horizon in dataset (#794) (f07bc5a)
- string operations on cells (#791) (4a17f76)
Bug Fixes
- handling of boolean columns in column statistics (#778) (f61cceb)
- sort x values of line plot (#782) (74d8649)
Special thanks to all contributors:
v0.25.0
0.25.0 (2024-05-15)
Features
-
$R^2$ metric for regressors - allow using tabular datasets in
predict
- implement tabular containers and all transformer (except
Discretizer
) with polars - move NN converters and layers to separate packages (#759) (c6a4073)
- remove deprecated methods from tabular containers
- specify partial order in label encoder (#763) (6fbe537), closes #639
- specify value separator when reading CSV file
- standalone metrics package
Special thanks to all contributors:
v0.24.0
0.24.0 (2024-05-09)
This release features completely rewritten containers for tabular data (currently experimental). They use the extremely fast polars library as their backend. Together with a drastically more efficient implementation of our own interface, operations on tabular data are now as fast as they should be.
Previously, even operations on small tables (10000 rows x 50 columns) took very long, as this comparison of Table
methods shows:
method | old (s) | new (s) | speedup (factor) |
---|---|---|---|
remove_duplicate_rows |
0.25474 | 0.01306 | 19.5 |
remove_rows_with_missing_values |
0.25159 | 0.00946 | 26.6 |
remove_rows_with_outliers |
0.28816 | 0.01034 | 27.9 |
remove_rows |
2.69647 | 0.00242 | 1114.2 |
shuffle_rows |
0.24690 | 0.00204 | 121.0 |
slice_rows |
0.12313 | 0.00011 | 1119.4 |
sort_rows |
4.67574 | 0.00372 | 1256.9 |
split_rows |
0.24764 | 0.00219 | 113.1 |
transform_column |
2.89572 | 0.00030 | 9652.4 ❗ |
You can find a full list of changes below. Special thanks to all contributors:
Features
Column.plot_histogram()
usingTable.plot_histograms
for consistent results (#726) (576492c)Regressor.summarize_metrics
andClassifier.summarize_metrics
(#729) (1cc14b1), closes #713- Add
ImageDataset
and Layer for ConvolutionalNeuralNetworks (#645) (5b6d219), closes #579 #580 #581 - added load_percentage parameter to ImageList.from_files to load a subset of the given files (#739) (0564b52), closes #736
- added rnn layer and TimeSeries conversion (#615) (6cad203), closes #614 #648 #656 #601
- Basic implementation of cell with polars (#734) (004630b), closes #712
- deprecate
Table.add_column
andTable.add_row
(#723) (5dd9d02), closes #722 - deprecated
Table.from_excel_file
andTable.to_excel_file
(#728) (c89e0bf), closes #727 - Larger histogram plot if table only has one column (#716) (31ffd12)
- polars implementation of a column (#738) (732aa48), closes #712
- polars implementation of a row (#733) (ff627f6), closes #712
- polars implementation of table (#744) (fc49895), closes #638 #641 #649 #712
- regularization for decision trees and random forests (#730) (102de2d), closes #700
- Remove device information in image class (#735) (d783caa), closes #524
- return fitted transformer and transformed table from
fit_and_transform
(#724) (2960d35), closes #613
Bug Fixes
Performance Improvements
v0.23.0
v0.22.1
v0.22.0
0.22.0 (2024-05-01)
Features
is_fitted
is now always a property (#662) (b1db881), closes #586- add
Column.missing_value_count
(#682) (f084916), closes #642 - Add
InputConversion
&OutputConversion
for nn interface (#625) (fd723f7), closes #621 - Add hash,eq and sizeof in ForwardLayer (#634) (72f7fde), closes #633
- allow using tables that already contain target for prediction (#687) (e9f1cfb), closes #636
- callback
Row.sort_columns
takes four parameters instead of two tuples (#683) (9c3e3de), closes #584 - rename
group_rows_by
inTable
togroup_rows
(#661) (c1644b7), closes #611 - rename
number_of_column
inRow
tonumber_of_columns
(#660) (0a08296), closes #646 - rework
TaggedTable
(#680) (db2b613), closes #647 - show missing value count/ratio in summarized statistics (#684) (74b8a35), closes #619
- specify
extras
instead offeatures
into_tabular_dataset
(#685) (841657f), closes #623
Bug Fixes
Performance Improvements
v0.21.0
0.21.0 (2024-04-17)
Features
- add ARIMA model (#577) (8b9c7a9), closes #570
- Add ImageList class (#534) (3cb74a2), closes #528 #599 #600
- more hash, sizeof and eq implementations (#609) (2bc0b0a)
Performance Improvements
- Add special case to
Table.add_rows
to increase performance (#608) (ffb8304), closes #606 - improve performance of model & forward layer (#616) (e856cd5), closes #610
- lazily import our modules and external libraries (#624) (20fc313)
- treat Tables specially when calling add_rows (#606) (e555b85), closes #575
v0.20.0
0.20.0 (2024-04-03)
Features
- add deterministic hash methods to all types (#573) (f6a3ca7)
- add fnn functionality (#529) (ce53153), closes #522
- add suffixes to models to indicate their task (#588) (d490dee)
- added lag_plot (#548) (0fb38d2), closes #519
- added normal plot for time series (#550) (dbdf11e), closes #549
- when using from table to time series feature must be given (#572) (ca23f0f), closes #571