v0.14.2

EvoTrees v0.14.2

Diff since v0.14.1

Merged pull requests:

• fix docs & add _mle suffix names for gaussian / logistic losses (#197) (@jeremiedb)

v0.14.1

EvoTrees v0.14.1

Diff since v0.14.0

Merged pull requests:

• Fix reproducibility (#196) (@jeremiedb)

v0.14.0

EvoTrees v0.14.0

Key change: rowsampling has been redisigned. Instead of returning a fixed number of candidate observation for each tree, each observation is randomly selected. The results in sampled num observations to be ~Binomial(num_obs, rowsample)

Diff since v0.13.1

Merged pull requests:

• Perf sample (#194) (@jeremiedb)

v0.13.1

EvoTrees v0.13.1

Diff since v0.13.0

Merged pull requests:

• simplify gain calculations (#193) (@jeremiedb)

v0.13.0

EvoTrees v0.13.0

(Multi)Classification / softmax now supported on GPU through EvoClassifier(device = "gpu", ...)
No more export of grow_tree! and predict.
For prediction / inference, the recommended to approach is (m::EvoTree)(x).
The fitted model struct is now EvoTree / EvoTreeGPU. (previously was GBTree / GBTreeGPU).
fit_evotree can now return the tracking of the metrics throughout the iterations by using return_logger = true as kwarg. Example:

m, logger = fit_evotree(config; x_train, y_train, x_eval, y_eval, metric=:logloss, print_every_n=50, early_stopping_rounds=100, return_logger = true)

Diff since v0.12.5

Closed issues:

• Feature request: support EvoTreeClassifier(loss = Softmax()) on the GPU (#139)
• MLJModelInterface.fit does not accept tables? (#169)
• Plot loss curve history for train and eval data set (#178)
• GPU for classifiers (#185)
• GPU saved model not possible open without CUDA (#189)

Merged pull requests:

• Logger (#183) (@jeremiedb)
• Add ntree_limit for predict (#184) (@Moelf)
• massively reduce CPU allocation (#186) (@Moelf)
• Dev (#187) (@jeremiedb)
• Gpu upgrade (#188) (@jeremiedb)
• Dev (#190) (@jeremiedb)
• Add generic MLJ interface tests (#191) (@ablaom)

v0.12.5

EvoTrees v0.12.5

Diff since v0.12.4

Closed issues:

• Perf improvements (#105)
• Feature Request (#161)
• how to free gpu memory after training with MLJ interface (#171)

Merged pull requests:

• Fix MLJ metadata declarations so that new docstrings get written to the MLJ model registry (#180) (@ablaom)
• Add weight support for MLJ & docs fix (#181) (@jeremiedb)
• threaded gradients (#182) (@jeremiedb)

v0.12.4

EvoTrees v0.12.4

Diff since v0.12.3

Merged pull requests:

• fix GC GPU & docs typo (#176) (@jeremiedb)

v0.12.3

EvoTrees v0.12.3

Diff since v0.12.2

Merged pull requests:

• print initial metric only if verbose > 0. Fix for #174 (#175) (@jeremiedb)

v0.12.2

EvoTrees v0.12.2

Diff since v0.12.1

Merged pull requests:

• Dev (#173) (@jeremiedb)

v0.12.1

EvoTrees v0.12.1

Diff since v0.12.0

Closed issues:

• Create MLJ-compliant doc strings (#145)
• Feature Importance Doesn't Allow Feature Names (#170)

Merged pull requests:

• Jdb/logistic (#172) (@jeremiedb)