tpot2

tpot2 copied to clipboard

Replace remaining calls to legacy `np.random.seed()`

1

Thanks for all of your work on TPOT! I and my colleagues use it very often for a variety of synthetic biology research projects. I'd like to ask that TPOT...

chimaerase

Population class logging is inefficient and needs to be optimized.

1

The population class stores its logs in a pandas data frame. This has very slow appends, which get slower the more items are in the data frame. This becomes a...

perib

Currently, the configuration files are designed as optuna-compatible functions, but this adds boilerplate (such as f"_{name}".) And maybe confusing and hard to understand. It would help with clarity if we...

perib

Fix miscellaneous invalid hyperparameter options in the configuration dictionary. And add n_jobs=1 when possible.

2

I went through some logs to check for classifiers, transformers, and selectors issues. Those seem to not generate invalid hyperparameter combinations. I have not checked regression yet. It would also...

perib

Benchmarks available?

3

Been a while since I used Tpot in a project, and would like to try out the V2 version in a subsequent upcoming project. Any benchmarks available to date comparing...

Et9797

We may need to differentiate between selection methods with possible replacement (for matching parents) and those without replacement (such as select best, used for cutting the population down) Should we...

perib

TPOT2 is missing the cuML [classifiers ](https://github.com/EpistasisLab/tpot/blob/master/tpot/config/classifier_cuml.py) and [regressors ](https://github.com/EpistasisLab/tpot/blob/master/tpot/config/regressor_cuml.py) found in TPOT1. These could be created and given their own category in [this function](https://github.com/EpistasisLab/tpot2/blob/main/tpot2/tpot_estimator/estimator_utils.py#L24)

perib

right now `population_size` is the number of generated offspring. `survival_percentage`*`population_size` is the number of parents to keep each generation. `survival_percentage` was developed with successive halving in mind, with the assumption...

perib

Currently, lambda and partial functions throw errors when used as custom objective functions if the user does not manually define a name for the function.

perib

A meta-learning model could be used to predict optimal parameters for the evolutionary algorithm given a dataset. For an example, see Auto-Sklearn [paper](https://automl.github.io/auto-sklearn/master/)

perib