- Ignore unamed experiment configuration file backups;
- Factorize default dataset loading parameters;
- Add missing return_X_y in basic dataset loaders.

Finish to fix sklearn dataset fetchers. Keep 20newsgroups_vectorized instead of non vectorized version that is the preprocessed version of this text dataset. Remove kddcup99 dataset.

Fix most of the missing dataset fetchers. Remainings: 20newsgroups, 20newsgroups_vectorized, kddcup99

- Update TODO list.

- Move bolsonaro imports on top of the file in compute_hyperparam file;
- Move change_binary_func_load to utils file.

When training, look if there is bayesian search results, if yes use this. Exception: forest_size use the one given by parser if applicable

Big changes: Create intermediate classes OMPForest and SingleOmpForest for code factoring: share code between OmpForestRegressor and OmpForestBinaryClassifer. Remove set_wweights and set_forest which are not relevant anymore. load function from model_factory isn't trustfull now: raises an error. TODO: multiclass classifier

Introduce the notion of "stage" in the experiments (see TODO comment in compute_results.py file. Add example of experiment configuration file tree in experiments/boston/stage3 for stage 3 (unoptimized parameters due to missing stage 1 and 2 results).

- Add experiment_configuration parameter to run an experiment from a json configuration file. If the experiment configuration are commnig from the arguments, save it to a file to keep trace of it;
- Add few comments in train.py.

- Use as much CPU as possible when training a random forest regressor.

Replace use_dev_subset by subsets_used parameter, in order to specify more clearly which combination of train dev to used to train the forest and OMP

- Add a TODO for the subset in train
- Remove an useless new line

- Save and load json files from __dict__.

- Fix bug of resolve_experiment_id after 10 exp ids;
- Display the new experiment id at the beginning of the training;
- For now there's only a simple losses plot in compute_results.

- Update requirements packages.

- In compute_results, add loadenv, load raw results, and update experiment_ids so that it's possible to specify a list of experiments ids. The default behavior is to load all exp ids;
- Fix normalization option in train.py. By default it normalizes D, and it doesn't when specify wo_normalization option;
- Fix logger.warn to logger.warning in train.py
- Replace the dumping of result in trainer.py by a dedicated class to save and load the trained model and training metadatas: model_raw_results.py;
- Rename too long func DatasetLoader.load_from_name to DatasetLoader.load;
- Add loading functions in dataset_parameters and model_parameters;
- Set console logging level to INFO to summarize the most important console logs;
- Add a load function in model_factory;
- In omp_forest_regressor, move private funcs to the bottom of the file.

TODO: compute the plot from the loaded raw results in compute_results file.

- Add train_on_subset option to specify on which subset the model will be trained (either train or dev);
- find_dotenv() only working by specifying the example env on my machine?
- Add the seeds option to specify the seed(s) to use, and remove the use_random_seed, because it's obv if random_seed_number is used;
- Use a logger in train.py instead of prints.