Resolve "Experiment pipeline"

Closes #12 (closed)

added 1 commit

• d70fad15 - Fix most of the missing dataset fetchers. Remainings: 20newsgroups,...

Compare with previous version

added 1 commit

• 967742a6 - Finish to fix sklearn dataset fetchers. Keep 20newsgroups_vectorized instead...

Compare with previous version

added 1 commit

• f61314f3 - - Add standard dataset scaling for dataset normalization;

Compare with previous version

added 1 commit

• 254e846e - By default, compute the hyperparams of the whole dataset list.

Compare with previous version

added 1 commit

• e319343d - Compute best hyperparameters over k seeds, and select the most commons.

Compare with previous version

added 1 commit

• 7494dbc4 - Rename 'use_different_seed_number' parameter to 'use_variable_seed_number'

Compare with previous version

added 1 commit

• 39d4540f - Remove useless Real import

Compare with previous version

added 1 commit

• 31d12659 - Fix few typos in new compute_hyperparams version.

Compare with previous version

added 3 commits

• 11017545 - - Replace the futures concurrence by joblib Parallel (and add optional tqdm progress bar);
• 880ff78f - - Add an option to not use the best hyperparameters file;
• fd6dbc7b - POC of possible wrong way to compute best hyperparams. Are there the best only...

Compare with previous version

added 1 commit

• 6a6cf747 - - Reduce the extracted forest sizes upper bound and number because OMP seems...

Compare with previous version

added 3 commits

• 649b4e64 - - Finish first stable version of stage1 plotting;
• e07b2b53 - Fix path issue in second plot of stage1
• 7330792e - Set dataset_name arg in compute_results to required (tmp)

Compare with previous version

added 1 commit

• b56e4254 - Hardcoded metric functions for comparison

Compare with previous version

added 1 commit

• 6239030f - Add requirements for conda

Compare with previous version

added 1 commit

• 3a2ec5cb - Fix score func in trainer

Compare with previous version

added 1 commit

• 2df5560f - Update plot in compute results to have two axis instead of two plots.

Compare with previous version

added 1 commit

• 76a58da9 - Len de len :x

Compare with previous version

added 2 commits

• 51ba8a0e - - Unignore results;
• e9d1cefa - Add missing files in previous commit...

Compare with previous version

added 1 commit

• 628fba24 - Fix missing title in stage1 loss figure (still on the left; TODO: center the title).

Compare with previous version

added 1 commit

• 17d3addc - - The trainer can now specify which metrics to use;

Compare with previous version

added 1 commit

• 58061ea4 - - Add command lines for stage2 experiments;

Compare with previous version