Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix tests + add titanic database #2

Merged
merged 6 commits into from
Jun 14, 2019
Merged

Fix tests + add titanic database #2

merged 6 commits into from
Jun 14, 2019

Conversation

gfournier
Copy link
Member

No description provided.

@gfournier
Copy link
Member Author

@LionelMassoulard committed Titanic dataset, source is here: https://github.com/gfournier/aikit-datasets/releases/tag/titanic-1.0.0

gfournier
gfournier commented Jun 7, 2019
View reviewed changes
aikit/tools/helper_functions.py
@@ -357,7 +357,7 @@ def clean_column(s):

r = s.strip().lower()
r = re.sub(r"[?\(\)/\[\]\\]", "", r)
r = re.sub("[:' \-\.\n]", "_", r)
r = re.sub(r"[:' \-\.\n]", "_", r)
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Add test for this method

@gfournier
Copy link
Member Author

Add longtest marker to test_TargetEncoderClassifierEntropy1

@gfournier gfournier mentioned this pull request Jun 7, 2019
Closed
@gfournier gfournier force-pushed the wip_fix_travis_pytest_launch branch from fb2fd8b to 5833f65 Compare June 7, 2019 19:21
LionelMassoulard
LionelMassoulard reviewed Jun 7, 2019
View reviewed changes
tests/datasets/test_datasets.py

res = load_dataset(name)
res = load_dataset(name, cache_dir=tempdir)

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think pytest has a fixture to for tmpdir
https://docs.pytest.org/en/latest/tmpdir.html

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

e2729f3

@LionelMassoulard
Copy link
Collaborator

@LionelMassoulard committed Titanic dataset, source is here: https://github.com/gfournier/aikit-datasets/releases/tag/titanic-1.0.0

there is a weird "._titanic.csv" file in the archive

@gfournier gfournier changed the title (wip) Fix pytest launch Fix test + add titanic database Jun 7, 2019
@gfournier gfournier changed the title Fix test + add titanic database Fix tests + add titanic database Jun 7, 2019
@gfournier
Copy link
Member Author

Green tests :)
With LightGBM, Gensim, NLTK, Graphviz

@LionelMassoulard
Copy link
Collaborator

Green tests :)
With LightGBM, Gensim, NLTK, Graphviz

It works on my PC as well. I think the last thing is to update the notebooks since the name of the columns has changed + new columns.
(By I'm not sur having a column with a dot is a good idea "home**.**test")

@gfournier gfournier reopened this Jun 8, 2019
@gfournier gfournier merged commit 5044b8e into societe-generale:master Jun 14, 2019
LionelMassoulard pushed a commit to LionelMassoulard/aikit that referenced this pull request Jun 15, 2019
@gfournier @LionelMassoulard
Fix tests + add titanic database (societe-generale#2)
cf67602 
* Fix pytest launch

* Add Titanic dataset

* Fix failing tests

* Add NLTK stopwords to Travis config

* Doc + use tmpdir fixture from pytest

* Update notebooks
LionelMassoulard added a commit that referenced this pull request Jun 22, 2019
@LionelMassoulard
Make groups available in scorer (#9)
98b2aba 
* Fix pytest launch

* Add Titanic dataset

* Fix failing tests

* Add NLTK stopwords to Travis config

* Doc + use tmpdir fixture from pytest

* Update notebooks

* Port for p37 sk21 (#1)

* update to work in sk21
* fix n_estimators and seed for RandomForest
* conditionnal tests if libraries not installed
* fix doc typos
* removed from doc while the blender is put in the lib
* fix seeds
* improve README.md

* Fix tests + add titanic database (#2)

* Fix pytest launch

* Add Titanic dataset

* Fix failing tests

* Add NLTK stopwords to Travis config

* Doc + use tmpdir fixture from pytest

* Update notebooks

* Update README.md with doc link (#7)

* Update README.md with doc link

+ some formatting fixes

* Update README.md

fix typos

* add a few tests to clean_column

* move "function_has_named_arguments" + add test with a functor

* create "GroupProbaScorer" : scorer that can access "groups" info

* modify cross validation to allow scorer to use a groups.
gfournier pushed a commit that referenced this pull request Aug 7, 2019
Block Search + other (#2)
33f2cd3 
* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test
gfournier pushed a commit that referenced this pull request Aug 7, 2019
@gfournier
Block Search + other (#2)
e28b813 
* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test
gfournier added a commit that referenced this pull request Aug 7, 2019
@gfournier
Integrate Jun-Jul developments (#14)
4f40fe3 
* Bump version to 0.1.0

* Change output type vectorizer (#1)

* change setup

* change default output type of countvectorizer to bet in32

* change dtype to numerical encoder as well + tests

* add output type test on NumImputer

* fix bug NumericalEncoder when new column (#4)

* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* Graph pipeline subgraph from dev (#3)

* fix casting bug + test on filter/map function on dicos

* add function to retrieve 2-uple list of edges from generic tuple edges

* fix bug on DebugPassThrough

* add 'get_subpipeline' methods to create sub GraphPipeline from a given GraphPipeline

* add docstring get_subpipeline

* Fix numerical encoder max_cum_proba (#6)

* Fix bug automl group (#5)

* allow reload of groups

*  * add average_precision default transformation

 * go back to default transformation if unknown

* return dataframe in command

* Fix dataset load from SG premises

* Fix dummy encoding type in NumericalEncoder
gfournier pushed a commit that referenced this pull request Oct 1, 2019
Dispatch groups (#7)
1b8459d 
* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* move 'function_has_named_argument' from .transformers.model_wrapper to .tools.helper_functions

* cleanning

* dispatch and split the groups variable to the estimator

* add groups to methods + dispatch it to estimators within the pipeline

* test on cross validation and pipeline to check the passing of groups

* remove useless import

* remove useless

* fix X -> lastX

* debug help

* fix after merge

* make sur benchmark can be computed

* input np.inf  as well as np.nan

* spaces

* don't split and tokenize if not needed

* new tests auto-ml, when only numerical values

* allow scoring to return multiple values

* allow cross_validation to be in Parallel

# Conflicts:
#	aikit/cross_validation.py

* add a custom CV for groups

*  * froze init param

 * allow additionnal function to be computed

* read additionnal results

* allow guiding to be done on an "addtionnal metric"

* typo

* add name of excel print

* test if name of columns has change
gfournier pushed a commit that referenced this pull request Oct 1, 2019
Ml graph improve (#8)
ead9e0e 
* * new helpers function (merge node and subbranch search)

* fix ordering in graph from edges

* * generalize the notion of model graph

* change name representation

* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* fix type : TransformToBlockManager

* add number of output utils function

* spaces

* new tests with impossible graphs

* fix merged

* fix notebook error

* add list test

* remove useless import

* spaces

* fix docstring

* merge 2 loops

* remove duplicate edge
gfournier pushed a commit that referenced this pull request Oct 1, 2019
@gfournier
Dispatch groups (#7)
ab3428d 
* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* move 'function_has_named_argument' from .transformers.model_wrapper to .tools.helper_functions

* cleanning

* dispatch and split the groups variable to the estimator

* add groups to methods + dispatch it to estimators within the pipeline

* test on cross validation and pipeline to check the passing of groups

* remove useless import

* remove useless

* fix X -> lastX

* debug help

* fix after merge

* make sur benchmark can be computed

* input np.inf  as well as np.nan

* spaces

* don't split and tokenize if not needed

* new tests auto-ml, when only numerical values

* allow scoring to return multiple values

* allow cross_validation to be in Parallel

# Conflicts:
#	aikit/cross_validation.py

* add a custom CV for groups

*  * froze init param

 * allow additionnal function to be computed

* read additionnal results

* allow guiding to be done on an "addtionnal metric"

* typo

* add name of excel print

* test if name of columns has change
gfournier pushed a commit that referenced this pull request Oct 1, 2019
@gfournier
Ml graph improve (#8)
0caa005 
* * new helpers function (merge node and subbranch search)

* fix ordering in graph from edges

* * generalize the notion of model graph

* change name representation

* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* fix type : TransformToBlockManager

* add number of output utils function

* spaces

* new tests with impossible graphs

* fix merged

* fix notebook error

* add list test

* remove useless import

* spaces

* fix docstring

* merge 2 loops

* remove duplicate edge
gfournier added a commit that referenced this pull request Oct 1, 2019
@gfournier
Integrate Aug-Sep developments (#20)
8b0498c 
* Bump version to 0.1.1

* Fix bug automl block search (#10)

* fix bug when no elements to iterator on

* remove useless space

* Categorical handling (WIP) (#9)

* add failing test for categorie

* - add a function that can replace categorical columns by object columns

- recognize 'category' as a CAT type of variable

* ajoute de get ride of categories

modifications des transfo numericalencoder et targetencoder
 ajout d un test de guess_type_of_variables

* - add a get_rid_of_categories in the fit_transform of targetencoder

- add test of targetencoder with categorical dtype
- add test of numericalencoder with categorical dtype

* modif de test_guesss_type_of_variable

* ajout d'un test permettant de vérifier que le numerical encoder ne transforme pas les colonnes catégorielles ayant des int en colonnes numériques.

pour l'instant, le test fail

* modification du code pour que le numericalencoder et le targetencoder fonctionnent correctement

ajout de tests

* modifs prenant en compte les comments de la pull request

* remaining changes for the pull request

* clean commit

* Dispatch groups (#7)

* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* move 'function_has_named_argument' from .transformers.model_wrapper to .tools.helper_functions

* cleanning

* dispatch and split the groups variable to the estimator

* add groups to methods + dispatch it to estimators within the pipeline

* test on cross validation and pipeline to check the passing of groups

* remove useless import

* remove useless

* fix X -> lastX

* debug help

* fix after merge

* make sur benchmark can be computed

* input np.inf  as well as np.nan

* spaces

* don't split and tokenize if not needed

* new tests auto-ml, when only numerical values

* allow scoring to return multiple values

* allow cross_validation to be in Parallel

# Conflicts:
#	aikit/cross_validation.py

* add a custom CV for groups

*  * froze init param

 * allow additionnal function to be computed

* read additionnal results

* allow guiding to be done on an "addtionnal metric"

* typo

* add name of excel print

* test if name of columns has change

* Clean load (#12)

* remove config.json

* fix loading

* remove nltk addtional path

* accelerate code using map and dict (#13)

* accelerate code using map and dict

* accelerate concatenation code

* Update categories

* * fix test new columns name (#15)

* fix seed
* new test CdfScaler

* Ml graph improve (#8)

* * new helpers function (merge node and subbranch search)

* fix ordering in graph from edges

* * generalize the notion of model graph

* change name representation

* Block Search + other (#2)

* add make_pipeline function (works like sklearn)

* fix type "_if_fitted" -> "_already_fitted"

*  * add handling of columns_to_encode == "--object--" in target encoder

 * corresponding test

* add Numerical encoder test for "columns_to_encode == '--object--' "

* expose command argument parser outside, to be able to add new arguments.

* change WordVectorizer in char mod distributions

+ fix bug in HyperRangeBetaInt

* change default behavior : encode "columns_to_encode == '--object--' "

* remove 'bug' (double return)

* allow text preprocessors to concat their inputs

* add 'RandomTrainTestCv' and 'IndexTrainCv' cv-like object.

* same api as a regular cv object ...
* ... but only one split

* add 'use_for_block_search' attribute + filter models based on that

* * add block search iterator

* automl config : models_to_keep_block_search

* fix typo in test

* ignore Warning in test

* fix type : TransformToBlockManager

* add number of output utils function

* spaces

* new tests with impossible graphs

* fix merged

* fix notebook error

* add list test

* remove useless import

* spaces

* fix docstring

* merge 2 loops

* remove duplicate edge

* add a few ploting functions (#14)

* add a few ploting functions

* add assert

* bump version 0.1.2

* DEV bump version

* doc typo (#16)

* Add matplotlib, seaborn to test requirements

* Fixes on dataset load from public URL

* Fix dataset path load unit test
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@LionelMassoulard LionelMassoulard LionelMassoulard left review comments

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@gfournier @LionelMassoulard