- added merge of text columns. Now text columns can be combined into one (early fusion strategy)
- test_data_merge_texts was changed

Fix for #979

* Image updated

* Upd python version in publish and workflows

* Fix leftover python version 3.7

* Remove python version 3.7 in requirements

* Update links to the project

* Update package version

* Remove sklearn without version from requirements

* Fixes after rebase

* Minor fix

* Delete tuning docs

* Change versions to pass tests

* Minor

* Minor

* Move doc files to golem

* miinor fixes to docs

* Add document page for interaction with GOLEM

* Fix doc errors

* Add link to GOLEM package

- added doc for multimodal tasks
- fixed the links in notebooks.rst

* Del tests that fully moved to GOLEM

* WIP transfer tests

* Fixes after rebase

* Remove some tests moved to golem

* Remove some tests moved to golem

* Remove some tests moved to golem

* Remove some tests moved to golem

* Fixes after rebase

* Remove PrimaryNode and SecondaryNode

* Fix test_complex_search_space_tuning_correct

* Upd requirements.txt with GOLEM git

* WIP drop files used in GOLEM

* WIP renames/edits for golem

* Major transition of imports (with optimization) to GOLEM

* Tests/examples/cases transition of imports to GOLEM

* Test fixes after transition

* Fix metrics_objective.py deserialization

* fix req-s for setup.py

* Docs fixes

* Add import transition script

* WIP composer_requirements movement

* WIP GraphOptimizerParameters rename

* WIP GPGraphOptimizerParameters rename

* rename max_pipeline_fit_time to max_graph_fit_time

* fix integration test on timeout (num of gen-s)

* fix integration test on parallel cache files

* rebase fixes

* WIP fix custom model example

* fix tests after changes in golem

* Remove random optimizers, add RandomPipelineFactory

* drop outdated test (will be fixed in golem)

* Fix integration test on quality

* Update requirement on golem, fix name

* Upd requirements for new golem package name

* minor: extend backward compat history test

* remove and fix opt graph buldier for golem

* fixes after rebase for golem

* Add 'stable' branch for dependency on GOLEM

* Fix imports after rbase

---------
Co-authored-by: Lyubov Yamshchikova <43475193+YamLyubov@users.noreply.github.com>

Some changes that can increase efficiency of clstm algorithm. Also bug with composer's inability to use model during composing was solved.

But using such heavy model in composing still questionable due to low performance compared with it's weight (maybe it can be useful on some types of data and with small horizon but I can't say exactly). Maybe we can refactor model structure, advices are welcome.

* release 0.6.2

* intergration test fixes

* intergration test fixes: disable tuning in a long test for cache

* intergration test fixes: disable tuning in more tests

* intergration test fixes: less timeout for profiler test

* intergration test fixes: simpler tasks or parallel_cache_files test

* add PipelineNode class

* refactor all usages with new node type

* remove outdated test

* remove PrimaryNode and SecondaryNode classes

* simplify code using PipelineNode

* minor fix tests

* minors with docs

* add comments

* remove Node class

+ pipeline's new parameter and preprocessing class
+ pipeline's integrational tests for the new parameter
+ split preprocessing status param in supplementary_data into obligatory and optional
+ main api's integrational test for the new parameter
+ type/style/logic fixes

* Fix plot_pareto

* Revert changes in get_metrics

* Correct docstring

* Correct docstring

* Fix for tuner test

* returning iris dataset for classification tests

* set false for iris in tune test

---------
Co-authored-by: Andrey Stebenkov <a.stebenkov75@yandex.ru>

Hotfix for #1021 - Changed preset from auto to glm and increase tuning timeout for failed test

* allow main API to use PathLike objects

* edit Fedot.fit() docstring

* minor fixes

* redefine target in case ts from array

* move pd.read_csv to separate method for separate data; add `possible_idx_keywords` argument

* improve docs

* unify all from_csv functions

* add logger message

* rename 'ml_task' -> 'task'

* fix typos, readability, docstring

* fix `from_csv_time_series`

* fixes for indexing
 - prevent parsing index as date for time series for consistency;
 - try to define index column even outside "columns_to_use".

* fix `test_output_mode_full_probs`

* minor fix

* add tests for defining index

Adding `.tune()` method in main API which allow tunes pipelines after composing or if predefined was selected.

* first approximation

* minor

* add skip connections method

* add tests

* add OptGraphBuilder & GraphBuilder

* refactor PipelineBuilder

* minors

* optimize imports

* corner situation witj cycle graphs

* minor

* minor fix in docstring

* Stopping if even one fold failed
* Presets bug fixed
* More stable wor for short timeouts
* Various minor changes
* Sequatial mode without Joblib parallelization
* Tuner timeouts processing

* add stratification for holdout

* remove excessive spaces

* add tests

* Add example for kc2 dataset

* Add experiment script

* Add baseline to kc2 example

* Add requirements for example reproducibility

* Memory analytics implemented

`np.shape` was misused in `test_history_backward_compatibility`.

According to this PR, Seaborn is not compatible with NumPy 1.24.0 due to a bug in the latter:

https://github.com/mwaskom/seaborn/pull/3194

Furthermore, the bug is going to be fixed by numpy 1.24.1:

https://github.com/numpy/numpy/milestone/114

Thus, restricting the bugged NumPy version only.

* minors

* fix

* add const

* minor

* revert

* fixes: fix in adapter for restore of returned values

* fixes: Restricts ParetoFront size as it was becoming very large

* fixes: multi-objective fitness was maximised

* fix: multi-objective metrics not handled correctly in API

* Add integration test for multi obj opt

* Review fixes

* Fix zero div error in ParetoFront

* fix similarity function for just equlaity in generation_keeper.py

* fix bad plotting in one example

- Removed deprecated dataset "Boston"
- Simplified bad-written fixtures
- Added scikit-learn 1.2 compatibility

* save method description

* load method

* fixes

* add example

* fix doc structure

* saving examples

* headers fix

* minors

* minors

* minors

* minors #12345

* minor

* fixes after review

* minor

* clickable section

* clickable section

* reformat text

* fix load

* minors

* WIP

* WIP forecasting docs

* Add train_test_split for ts docs

* Add prediction section

* Add manual and multivariate example

* Remove ts_gapfilling

* Fix title

* Add explanations

* Change pipeline in example

* Add plots of the pipeline and rewrite prediction description

* Extend Prediction sample

* Add plots for Prediction section

* Add multimodal example

* Minor fix

* Minor changes

* Change pipeline plots

* Correct typos

* Refactor predict for ts to be in_sample

* Minor fixes

* Fix plot_prediction

* Minor fixes

* Correct in-sample forecasting

* Fix doc strings

* Fix in sample predict fot multi_ts data type

* Add test for multi_ts in-sample predict

* Add test for multimodal ts data

* Rename variable target in plot_forecast

* Add prediction_is_in_sample flag

* Fix tests

* Rename flag prediction_is_in_sample to _is_in_sample_prediction