EpistasisLab / tpot2

A Python Automated Machine Learning tool that optimizes machine learning pipelines using genetic programming.

https://epistasislab.github.io/tpot2/

EpistasisLab/tpot2 Issues

Add a destroyer function to properly close dask clients when the evolver or estimator instance is deleted.
Closed 4 months ago
Learning Machine
Closed 4 months ago1
Replace remaining calls to legacy `np.random.seed()`
Updated 6 months ago
Need to update the conda package to the latest version of TPOT2 (can this be automatic? )
Closed 7 months ago3
Population class logging is inefficient and needs to be optimized.
Updated 7 months ago1
Can we improve the configuration files to be easier to read and write?
Updated 8 months ago
Fix miscellaneous invalid hyperparameter options in the configuration dictionary. And add n_jobs=1 when possible.
Updated 8 months ago2
API : improve selector module organization
Updated 8 months ago
TPOT1 feature Parity: cuML configuration dictionaries
Updated 8 months ago
Rename survival_percentage to something clearer.
Updated 8 months ago
Automatically generate names for lambda/partial functions used as custom objective functions
Updated 8 months ago
Implement Meta Learner
Updated 8 months ago
Idea for a two step preprocessing pipeline
Updated 8 months ago
Add support for ordinal categorical columns
Updated 8 months ago
Better integrate Optuna Optimization
Updated 8 months ago
implement a callback function/class
Updated 8 months ago
Genetic Feature Selection Improvements
Updated 8 months ago
optimization of preprocessing pipeline
Updated 8 months ago
add support for regressors in the complexity objective function
Updated 8 months ago
This issue for TPOT1 brought up a good idea about outlier detection https://github.com/EpistasisLab/tpot/issues/1166
Updated 8 months ago
Add option for successive halving based on time or convergence rather than number of generations.
Updated 8 months ago
add in unit tests
Updated 8 months ago
TPOT1 Feature Parity: Write an export function that takes a graphpipeline and generates a .py file similar to TPOT1
Updated 8 months ago
TPOT1 Feature Parity: add "sample_weight" and "groups" parameters to the fit function in TPOTEstimator and the graphpipeline
Updated 8 months ago
Add successive halving algorithm to steady state
Updated 8 months ago
add a debugging tutorial
Updated 8 months ago
Avoid duplicate work for steady state and base evolvers. Have one evolver that can do both generational EA, but also allow for extra individuals to be generated on the fly to make use of idle cores.
Updated 8 months ago
Support custom initializations / manually defined initial pipelines -
Updated 8 months ago
rename baseevolver.optimize to .evolve? To avoid confusion with individual.optimize
Updated 8 months ago
allow individual modules to be specified by string for config dicts. For example ["LogisticRegression", "KNN"] etc.
Updated 8 months ago
Licensing text in source files
Updated 8 months ago
Make the cv early stop functions into standalone functions? Or remove?
Updated 8 months ago1
Test that TPOTEstimator, TPOTClassifier, TPOTRegressor have a) same params, b) meet the sklearn baseestimator conventions
Updated 8 months ago
shuffle cv folds every generation?
Updated 8 months ago
Idea: Preventing getting stuck with variable number of mutations
Updated 8 months ago
add a scorer that counts selected number of features
Updated 8 months ago
Idea for Successive Halving to avoid repeated re-evaluations of the same pipeline.
Updated 8 months ago
edit/consolidate default configuration dictionary
Updated 8 months ago
replace instances of get_pareto_front with get_pareto_frontier
Updated 8 months ago
Avoiding repeated work : Cache node models with node hashing or manually.
Updated 8 months ago
add covariate adjustment
Updated 8 months ago
function that generates a learning curve
Updated 8 months ago
better plotting for graph pipeline
Updated 8 months ago
better thread efficiency for cv prunning with thresholds
Updated 8 months ago
Memory caching with GraphPipeline may miss some nodes where the ordering on inputs happens to be different between two nodes. - potential solution is to make graphs canonical - there is a C package that can be imported into TPOT that may do this
Updated 8 months ago
FeatureSetSelector module needs to implement a mask to be consistent with sklearn
Updated 8 months ago
xgboost only supports int labels (0-n). This is inconsistent with other sklearn methods. But we may have to manually transform labels for classification (making sure to keep with baseestitor conventions with est.classes_ etc.)
Closed 9 months ago1
Error in Last Block of Tutorial 7
Closed 9 months ago2
Error in Tutorial 7
Closed 10 months ago2
Link GitHub Pages in README.md
Closed 10 months ago1