Changelog¶
- v0.16.0 June 5, 2020
- Testing Changes
Update tests for numpy v1.19.0 compatability (GH#1016)
Thanks to the following people for contributing to this release: @Alex-Monahan, @frances-h, @gsheni, @rwedge, @thehomebrewnerd
- v0.15.0 May 29, 2020
- Enhancements
Add
get_default_aggregation_primitives
andget_default_transform_primitives
(GH#945)Allow cutoff time dataframe columns to be in any order (GH#969, GH#995)
Add Age primitive, and make it a default transform primitive for DFS (GH#987)
Add
include_cutoff_time
arg - control whether data at cutoff times are included in feature calculations (GH#959)Allow
variables_types
to be referenced by theirtype_string
for theentity_from_dataframe
function (GH#988)
- Fixes
Fix errors with Equals and NotEquals primitives when comparing categoricals or different dtypes (GH#968)
Normalized type_strings of
Variable
classes so that thefind_variable_types
function produces a dictionary with a clear key to name transition (GH#982, GH#996)Remove pandas.datetime in test_calculate_feature_matrix due to deprecation (GH#998)
Thanks to the following people for contributing to this release: @ctduffy, @frances-h, @gsheni, @jeff-hernandez, @rightx2, @rwedge, @sebrahimi1988, @thehomebrewnerd, @tuethan1999
Breaking Changes
Calls to
featuretools.dfs
orfeaturetools.calculate_feature_matrix
that use a cutoff time dataframe, but do not label the time column with either the target entity time index variable name or astime
, will now result in anAttributeError
. Previously, the time column was selected to be the first column that was not the instance id column. With this update, the position of the column in the dataframe is no longer used to determine the time column. Now, both instance id columns and time columns in a cutoff time dataframe can be in any order as long as they are named properly.The
type_string
attributes of allVariable
subclasses are now a snake case conversion of their class names. This changes thetype_string
of theUnknown
,IPAddress
,EmailAddress
,SubRegionCode
,FilePath
,LatLong
, andZIPcode
classes. Old saved entitysets that used these variables may load incorrectly.
- v0.14.0 Apr 30, 2020
- Enhancements
ft.encode_features - use less memory for one-hot encoded columns (GH#876)
Thanks to the following people for contributing to this release: @frances-h, @gsheni, @jeff-hernandez, @rwedge
Breaking Changes
Using training windows in feature calculations can result in different values than previous versions. This was done to prevent consecutive training windows from overlapping by excluding data at the oldest point in time. For example, if we use a cutoff time at the first minute of the hour with a one hour training window, the first minute of the previous hour will no longer be included in the feature calculation.
- v0.13.4 Mar 27, 2020
Warning
The next non-bugfix release of Featuretools will not support Python 3.5
- Fixes
Fix ft.show_info() not displaying in Jupyter notebooks (GH#863)
- Testing Changes
Miscellaneous changes (GH#861)
Thanks to the following people for contributing to this release: @frances-h, @FreshLeaf8865, @jeff-hernandez, @rwedge, @thehomebrewnerd
- v0.13.3 Feb 28, 2020
- Fixes
Fix a connection closed error when using n_jobs (GH#853)
- Changes
Pin msgpack dependency for Python 3.5; remove dataframe from Dask dependency (GH#851)
- Documentation Changes
Update link to help documentation page in Github issue template (GH#855)
Thanks to the following people for contributing to this release: @frances-h, @rwedge
- v0.13.2 Jan 31, 2020
- Enhancements
Support for Pandas 1.0.0 (GH#844)
- Changes
Remove dependency on s3fs library for anonymous downloads from S3 (GH#825)
- Testing Changes
Added GitHub Action to automatically run performance tests (GH#840)
Thanks to the following people for contributing to this release: @frances-h, @rwedge
- v0.13.1 Dec 28, 2019
Thanks to the following people for contributing to this release: @jeffzi, @kmax12, @rwedge, @systemshift
- v0.13.0 Nov 30, 2019
- Enhancements
Added GitHub Action to auto upload releases to PyPI (GH#816)
- Fixes
Fix issue where some primitive options would not be applied (GH#807)
Fix issue with converting to pickle or parquet after adding interesting features (GH#798, GH#823)
Diff primitive now calculates using all available data (GH#824)
Prevent DFS from creating Identity Features of globally ignored variables (GH#819)
Thanks to the following people for contributing to this release: @frances-h, @jeff-hernandez, @rwedge, @systemshift
Breaking Changes
The libraries used for downloading or uploading from S3 or URLs are now optional and will no longer be installed by default. To use this functionality they will need to be installed separately.
The fix to how the Diff primitive is calculated may slow down the overall calculation time of feature lists that use this primitive.
- v0.12.0 Oct 31, 2019
Thanks to the following people for contributing to this release: @ablacke-ayx, @BoopBoopBeepBoop, @jeffzi, @kmax12, @rwedge, @thehomebrewnerd, @twdobson
- v0.11.0 Sep 30, 2019
Warning
The next non-bugfix release of Featuretools will not support Python 2
- Enhancements
Improve how files are copied and written (GH#721)
Add number of rows to graph in entityset.plot (GH#727)
Added support for pandas DateOffsets in DFS and Timedelta (GH#732)
Enable feature-specific top_n value using a dictionary in encode_features (GH#735)
Added progress_callback parameter to dfs() and calculate_feature_matrix() (GH#739, GH#745)
Enable specifying primitives on a per column or per entity basis (GH#748)
- Fixes
Fixed entity set deserialization (GH#720)
Added error message when DateTimeIndex is a variable but not set as the time_index (GH#723)
Fixed CumCount and other group-by transform primitives that take ID as input (GH#733, GH#754)
Fix progress bar undercounting (GH#743)
Updated training_window error assertion to only check against observations (GH#728)
Don’t delete the whole destination folder while saving entityset (GH#717)
- Changes
Raise warning and not error on schema version mismatch (GH#718)
Change feature calculation to return in order of instance ids provided (GH#676)
Removed time remaining from displayed progress bar in dfs() and calculate_feature_matrix() (GH#739)
Raise warning in normalize_entity() when time_index of base_entity has an invalid type (GH#749)
Remove toolz as a direct dependency (GH#755)
Allow boolean variable types to be used in the Multiply primitive (GH#756)
- Documentation Changes
Updated URL for Compose (GH#716)
Thanks to the following people for contributing to this release: @angela97lin, @chidauri, @christopherbunn, @frances-h, @jeff-hernandez, @kmax12, @MarcoGorelli, @rwedge, @thehomebrewnerd
Breaking Changes
Feature calculations will return in the order of instance ids provided instead of the order of time points instances are calculated at.
- v0.10.1 Aug 25, 2019
- Fixes
Fix serialized LatLong data being loaded as strings (GH#712)
- Documentation Changes
Fixed FAQ cell output (GH#710)
Thanks to the following people for contributing to this release: @gsheni, @rwedge
- v0.10.0 Aug 19, 2019
Warning
The next non-bugfix release of Featuretools will not support Python 2
- Enhancements
Give more frequent progress bar updates and update chunk size behavior (GH#631, GH#696)
Added drop_first as param in encode_features (GH#647)
Added support for stacking multi-output primitives (GH#679)
Generate transform features of direct features (GH#623)
Added serializing and deserializing from S3 and deserializing from URLs (GH#685)
Added nlp_primitives as an add-on library (GH#704)
Added AutoNormalize to Featuretools plugins (GH#699)
Added functionality for relative units (month/year) in Timedelta (GH#692)
Added categorical-encoding as an add-on library (GH#700)
- Fixes
Fix performance regression in DFS (GH#637)
Fix deserialization of feature relationship path (GH#665)
Set index after adding ancestor relationship variables (GH#668)
Fix user-supplied variable_types modification in Entity init (GH#675)
Don’t calculate dependencies of unnecessary features (GH#667)
Prevent normalize entity’s new entity having same index as base entity (GH#681)
Update variable type inference to better check for string values (GH#683)
- Changes
Moved dask, distributed imports (GH#634)
Thanks to the following people for contributing to this release: @alexjwang, @allisonportis, @ayushpatidar, @CJStadler, @ctduffy, @gsheni, @jeff-hernandez, @jeremyliweishih, @kmax12, @rwedge, @zhxt95,
- v0.9.1 July 3, 2019
- Fixes
Select columns of dataframe using a list (GH#615)
Change type of features calculated on Index features to Categorical (GH#602)
Filter dataframes through forward relationships (GH#625)
Specify Dask version in requirements for python 2 (GH#627)
Keep dataframe sorted by time during feature calculation (GH#626)
Fix bug in encode_features that created duplicate columns of features with multiple outputs (GH#622)
Thanks to the following people for contributing to this release: @CJStadler, @kmax12, @rwedge, @gsheni, @kkleidal, @ctduffy
- v0.9.0 June 19, 2019
- Enhancements
Add unit parameter to timesince primitives (GH#558)
Add ability to install optional add on libraries (GH#551)
Load and save features from open files and strings (GH#566)
Support custom variable types (GH#571)
Support entitysets which have multiple paths between two entities (GH#572, GH#544)
Added show_info function, more output information added to CLI featuretools info (GH#525)
- Fixes
Normalize_entity specifies error when ‘make_time_index’ is an invalid string (GH#550)
Schema version added for entityset serialization (GH#586)
Renamed features have names correctly serialized (GH#585)
Improved error message for index/time_index being the same column in normalize_entity and entity_from_dataframe (GH#583)
Removed unused variable in normalize entity (GH#589)
Change time since return type to numeric (GH#606)
Thanks to the following people for contributing to this release: @alexjwang, @allisonportis, @CJStadler, @ctduffy, @gsheni, @kmax12, @rwedge
- v0.8.0 May 17, 2019
Rename NUnique to NumUnique (GH#510)
Serialize features as JSON (GH#532)
Drop all variables at once in normalize_entity (GH#533)
Remove unnecessary sorting from normalize_entity (GH#535)
Features cache their names (GH#536)
Only calculate features for instances before cutoff (GH#523)
Remove all relative imports (GH#530)
Added FullName Variable Type (GH#506)
Add error message when target entity does not exist (GH#520)
New demo links (GH#542)
Remove duplicate features check in DFS (GH#538)
featuretools_primitives entry point expects list of primitive classes (GH#529)
Update ALL_VARIABLE_TYPES list (GH#526)
More Informative N Jobs Prints and Warnings (GH#511)
Update sklearn version requirements (GH#541)
Update Makefile (GH#519)
Remove unused parameter in Entity._handle_time (GH#524)
Remove build_ext code from setup.py (GH#513)
Documentation updates (GH#512, GH#514, GH#515, GH#521, GH#522, GH#527, GH#545)
Thanks to the following people for contributing to this release: @bphi, @CharlesBradshaw, @CJStadler, @glentennis, @gsheni, @kmax12, @rwedge
Breaking Changes
NUnique
has been renamed toNumUnique
.Previous behavior
from featuretools.primitives import NUnique
New behavior
from featuretools.primitives import NumUnique
- v0.7.1 Apr 24, 2019
Automatically generate feature name for controllable primitives (GH#481)
Primitive docstring updates (GH#489, GH#492, GH#494, GH#495)
Change primitive functions that returned strings to return functions (GH#499)
CLI customizable via entrypoints (GH#493)
Improve calculation of aggregation features on grandchildren (GH#479)
Refactor entrypoints to use decorator (GH#483)
Include doctests in testing suite (GH#491)
Documentation updates (GH#490)
Update how standard primitives are imported internally (GH#482)
Thanks to the following people for contributing to this release: @bukosabino, @CharlesBradshaw, @glentennis, @gsheni, @jeff-hernandez, @kmax12, @minkvsky, @rwedge, @thehomebrewnerd
- v0.7.0 Mar 29, 2019
Improve Entity Set Serialization (GH#361)
Support calling a primitive instance’s function directly (GH#461, GH#468)
Support other libraries extending featuretools functionality via entrypoints (GH#452)
Remove featuretools install command (GH#475)
Add commutative argument to SubtractNumeric and DivideNumeric primitives (GH#457)
Add FilePath variable_type (GH#470)
Add PhoneNumber, DateOfBirth, URL variable types (GH#447)
Generalize infer_variable_type, convert_variable_data and convert_all_variable_data methods (GH#423)
Thanks to the following people for contributing to this release: @bukosabino, @CharlesBradshaw, @ColCarroll, @glentennis, @grayskripko, @gsheni, @jeff-hernandez, @jrkinley, @kmax12, @RogerTangos, @rwedge
Breaking Changes
ft.dfs
now has agroupby_trans_primitives
parameter that DFS uses to automatically construct features that group by an ID column and then apply a transform primitive to search group. This change applies to the following primitives:CumSum
,CumCount
,CumMean
,CumMin
, andCumMax
.Previous behavior
ft.dfs(entityset=es, target_entity='customers', trans_primitives=["cum_mean"])
New behavior
ft.dfs(entityset=es, target_entity='customers', groupby_trans_primitives=["cum_mean"])
Related to the above change, cumulative transform features are now defined using a new feature class,
GroupByTransformFeature
.Previous behavior
ft.Feature([base_feature, groupby_feature], primitive=CumulativePrimitive)
New behavior
ft.Feature(base_feature, groupby=groupby_feature, primitive=CumulativePrimitive)
- v0.6.1 Feb 15, 2019
Cumulative primitives (GH#410)
Entity.query_by_values now preserves row order of underlying data (GH#428)
Implementing Country Code and Sub Region Codes as variable types (GH#430)
Added IPAddress and EmailAddress variable types (GH#426)
Install data and dependencies (GH#403)
Add TimeSinceFirst, fix TimeSinceLast (GH#388)
Allow user to pass in desired feature return types (GH#372)
Add new configuration object (GH#401)
Replace NUnique get_function (GH#434)
_calculate_idenity_features now only returns the features asked for, instead of the entire entity (GH#429)
Primitive function name uniqueness (GH#424)
Update NumCharacters and NumWords primitives (GH#419)
Change to zipcode rep, str for pandas (GH#418)
Remove pandas version upper bound (GH#408)
Make S3 dependencies optional (GH#404)
Check that agg_primitives and trans_primitives are right primitive type (GH#397)
Mean primitive changes (GH#395)
Fix transform stacking on multi-output aggregation (GH#394)
Fix list_primitives (GH#391)
Documentation updates (GH#400, GH#409, GH#415, GH#417, GH#420, GH#421, GH#422, GH#431)
Thanks to the following people for contributing to this release: @CharlesBradshaw, @csala, @floscha, @gsheni, @jxwolstenholme, @kmax12, @RogerTangos, @rwedge
- v0.6.0 Jan 30, 2018
Primitive refactor (GH#364)
Mean ignore NaNs (GH#379)
Plotting entitysets (GH#382)
Add seed features later in DFS process (GH#357)
Multiple output column features (GH#376)
Add ZipCode Variable Type (GH#367)
Add primitive.get_filepath and example of primitive loading data from external files (GH#380)
Transform primitives take series as input (GH#385)
Add modulo to override tests (GH#384)
Thanks to the following people for contributing to this release: @floscha, @gsheni, @kmax12, @RogerTangos, @rwedge
- v0.5.1 Dec 17, 2018
- v0.5.0 Dec 17, 2018
Add specific error for duplicate additional/copy_variables in normalize_entity (GH#348)
Removed EntitySet._import_from_dataframe (GH#346)
Removed time_index_reduce parameter (GH#344)
Allow installation of additional primitives (GH#326)
Fix DatetimeIndex variable conversion (GH#342)
Update Sklearn DFS Transformer (GH#343)
Clean up entity creation logic (GH#336)
remove casting to list in transform feature calculation (GH#330)
Fix sklearn wrapper (GH#335)
Add readme to pypi
Update conda docs after move to conda-forge (GH#334)
Add wrapper for scikit-learn Pipelines (GH#323)
Remove parse_date_cols parameter from EntitySet._import_from_dataframe (GH#333)
Thanks to the following people for contributing to this release: @bukosabino, @georgewambold, @gsheni, @jeff-hernandez, @kmax12, and @rwedge.
- v0.4.1 Nov 29, 2018
Resolve bug preventing using first column as index by default (GH#308)
Handle return type when creating features from Id variables (GH#318)
Make id an optional parameter of EntitySet constructor (GH#324)
Handle primitives with same function being applied to same column (GH#321)
Update requirements (GH#328)
Clean up DFS arguments (GH#319)
Clean up Pandas Backend (GH#302)
Update properties of cumulative transform primitives (GH#320)
Feature stability between versions documentation (GH#316)
Add download count to GitHub readme (GH#310)
Fixed #297 update tests to check error strings (GH#303)
Remove usage of fixtures in agg primitive tests (GH#325)
- v0.4.0 Oct 31, 2018
Remove ft.utils.gen_utils.getsize and make pympler a test requirement (GH#299)
Update requirements.txt (GH#298)
Refactor EntitySet.find_path(…) (GH#295)
Clean up unused methods (GH#293)
Remove unused parents property of Entity (GH#283)
Removed relationships parameter (GH#284)
Improve time index validation (GH#285)
Encode features with “unknown” class in categorical (GH#287)
Allow where clauses on direct features in Deep Feature Synthesis (GH#279)
Change to fullargsspec (GH#288)
Parallel verbose fixes (GH#282)
Update tests for python 3.7 (GH#277)
Check duplicate rows cutoff times (GH#276)
Load retail demo data using compressed file (GH#271)
- v0.3.1 Sept 28, 2018
Handling time rewrite (GH#245)
Update deep_feature_synthesis.py (GH#249)
Handling return type when creating features from DatetimeTimeIndex (GH#266)
Update retail.py (GH#259)
Improve Consistency of Transform Primitives (GH#236)
Update demo docstrings (GH#268)
Handle non-string column names (GH#255)
Clean up merging of aggregation primitives (GH#250)
Add tests for Entity methods (GH#262)
Handle no child data when calculating aggregation features with multiple arguments (GH#264)
Add is_string utils function (GH#260)
Update python versions to match docker container (GH#261)
Handle where clause when no child data (GH#258)
No longer cache demo csvs, remove config file (GH#257)
Avoid stacking “expanding” primitives (GH#238)
Use randomly generated names in retail csv (GH#233)
Update README.md (GH#243)
- v0.3.0 Aug 27, 2018
Improve performance of all feature calculations (GH#224)
Update agg primitives to use more efficient functions (GH#215)
Optimize metadata calculation (GH#229)
More robust handling when no data at a cutoff time (GH#234)
Workaround categorical merge (GH#231)
Switch which CSV is associated with which variable (GH#228)
Remove unused kwargs from query_by_values, filter_and_sort (GH#225)
Remove convert_links_to_integers (GH#219)
Add example of using Dask to parallelize to docs (GH#221)
- v0.2.2 Aug 20, 2018
Remove unnecessary check no related instances call and refactor (GH#209)
Improve memory usage through support for pandas categorical types (GH#196)
Bump minimum pandas version from 0.20.3 to 0.23.0 (GH#216)
Make primitive lookup case insensitive (GH#213)
Use capital name (GH#211)
Set class name for Min (GH#206)
Remove
variable_types
from normalize entity (GH#205)Handle parquet serialization with last time index (GH#204)
Reset index of cutoff times in calculate feature matrix (GH#198)
Check argument types for .normalize_entity (GH#195)
Type checking ignore entities. (GH#193)
- v0.2.1 July 2, 2018
- v0.2.0 June 22, 2018
Multiprocessing (GH#170)
Handle unicode encoding in repr throughout Featuretools (GH#161)
Clean up EntitySet class (GH#145)
Add support for building and uploading conda package (GH#167)
Parquet serialization (GH#152)
Remove variable stats (GH#171)
Make sure index variable comes first (GH#168)
No last time index update on normalize (GH#169)
Remove list of times as on option for cutoff_time in calculate_feature_matrix (GH#165)
Config does error checking to see if it can write to disk (GH#162)
- v0.1.21 May 30, 2018
No EntitySet required in loading/saving features (GH#141)
Use s3 demo csv with better column names (GH#139)
more reasonable start parameter (GH#149)
add issue template (GH#133)
Update documentation after recent changes / removals (GH#157)
Rename demo retail csv file (GH#148)
Add names for binary (GH#142)
EntitySet repr to use get_name rather than id (GH#134)
Ensure config dir is writable (GH#135)
- v0.1.20 Apr 13, 2018
Primitives as strings in DFS parameters (GH#129)
Integer time index bugfixes (GH#128)
Add make_temporal_cutoffs utility function (GH#126)
Show all entities, switch shape display to row/col (GH#124)
Improved chunking when calculating feature matrices (GH#121)
fixed num characters nan fix (GH#118)
modify ignore_variables docstring (GH#117)
- v0.1.19 Mar 21, 2018
More descriptive DFS progress bar (GH#69)
Convert text variable to string before NumWords (GH#106)
EntitySet.concat() reindexes relationships (GH#96)
Keep non-feature columns when encoding feature matrix (GH#111)
Uses full entity update for dependencies of uses_full_entity features (GH#110)
Update column names in retail demo (GH#104)
Handle Transform features that need access to all values of entity (GH#91)
- v0.1.18 Feb 27, 2018
- v0.1.17 Jan 18, 2018
LatLong type (GH#57)
Last time index fixes (GH#70)
Make median agg primitives ignore nans by default (GH#61)
Remove Python 3.4 support (GH#64)
Change normalize_entity to update secondary_time_index (GH#59)
Unpin requirements (GH#53)
associative -> commutative (GH#56)
Add Words and Chars primitives (GH#51)
- v0.1.16 Dec 19, 2017
- v0.1.15 Dec 18, 2017
- v0.1.14 November 20, 2017
- v0.1.13 November 1, 2017
Add MANIFEST.in (GH#26)
- v0.1.11 October 31, 2017
Package linting (GH#7)
Custom primitive creation functions (GH#13)
Split requirements to separate files and pin to latest versions (GH#15)
Select low information features (GH#18)
Fix docs typos (GH#19)
Fixed Diff primitive for rare nan case (GH#21)
added some mising doc strings (GH#23)
Trend fix (GH#22)
Remove as_dir=False option from EntitySet.to_pickle() (GH#20)
Entity Normalization Preserves Types of Copy & Additional Variables (GH#25)
- v0.1.10 October 12, 2017
NumTrue primitive added and docstring of other primitives updated (GH#11)
fixed hash issue with same base features (GH#8)
Head fix (GH#9)
Fix training window (GH#10)
Add associative attribute to primitives (GH#3)
Add status badges, fix license in setup.py (GH#1)
fixed head printout and flight demo index (GH#2)
- v0.1.9 September 8, 2017
Documentation improvements
New
featuretools.demo.load_mock_customer
function
- v0.1.8 September 1, 2017
Bug fixes
Added
Percentile
transform primitive
- v0.1.7 August 17, 2017
Performance improvements for approximate in
calculate_feature_matrix
anddfs
Added
Week
transform primitive
- v0.1.6 July 26, 2017
Added
load_features
andsave_features
to persist and reload featuresAdded save_progress argument to
calculate_feature_matrix
Added approximate parameter to
calculate_feature_matrix
anddfs
Added
load_flight
to ft.demo
- v0.1.5 July 11, 2017
Windows support
- v0.1.3 July 10, 2017
Renamed feature submodule to primitives
Renamed prediction_entity arguments to target_entity
Added training_window parameter to
calculate_feature_matrix
- v0.1.2 July 3rd, 2017
Initial release