package orf

  1. Overview
  2. Docs
OCaml Random Forests

Install

Dune Dependency

Authors

Maintainers

Sources

v1.0.0.tar.gz
sha256=6dd5e7231fd11ffd9fdca3e58eeb3f1fde76229f655277f0e651165f82bdd56a
md5=1067c58787df67df21f3d6b520a5c71c

Description

Random Forests (RFs) can do classification or regression modeling.

Random Forests are one of the workhorse of modern machine learning. Especially, they cannot over-fit to the training set, are fast to train, predict fast, parallelize well and give you a reasonable model even without optimizing the model's default hyper-parameters. In other words, it is hard to shoot yourself in the foot while training or exploiting a Random Forests model. In comparison, with deep neural networks it is very easy to shoot yourself in the foot.

Using out of bag (OOB) samples, you can even get an idea of a RFs performance, without the need for a held out (test) data-set.

Their only drawback is that RFs, being an ensemble model, cannot predict values which are outside of the training set range of values (this is a serious limitation in case you are trying to optimize or minimize something in order to discover outliers, compared to your training set samples).

For the moment, this implementation only consider a sparse vector of integers as features. i.e. categorical variables will need to be one-hot-encoded. For classification, the dependent variable must be an integer (encoding a class label). For regression, the dependent variable must be a float.

Bibliography

Breiman, Leo. (1996). Bagging Predictors. Machine learning, 24(2), 123-140.

Breiman, Leo. (2001). Random Forests. Machine learning, 45(1), 5-32.

Geurts, P., Ernst, D., & Wehenkel, L. (2006). Extremely Randomized Trees. Machine learning, 63(1), 3-42.

Published: 21 Jun 2021

README

ORF: OCaml Random Forests

Random Forests (RFs) are one of the workhorse of modern machine learning. Especially, they cannot over-fit to the training set, are fast to train, predict fast, parallelize well and give you a reasonable model even without optimizing the model's default hyper-parameters. In other words, it is hard to shoot yourself in the foot while training or exploiting a Random Forests model. In comparison, with deep neural networks it is very easy to shoot yourself in the foot.

Using out of bag (OOB) samples, you can even get an idea of a RFs performance, without the need for a held out (test) dataset.

Their only drawback is that RFs, being an ensemble model, cannot predict values which are outside of the training set range of values (this is a serious limitation in case you are trying to optimize or minimize something in order to discover outliers, compared to your training set samples).

For the moment, this implementation will only consider a sparse vector of integers as features. i.e. categorical variables will need to be one-hot-encoded.

Bibliography

Breiman, Leo. (1996). "Bagging Predictors". Machine learning, 24(2), 123-140.

Breiman, Leo. (2001). "Random Forests". Machine learning, 45(1), 5-32.

Geurts, P., Ernst, D., & Wehenkel, L. (2006). "Extremely Randomized Trees". Machine learning, 63(1), 3-42.

Dependencies (8)

  1. line_oriented
  2. parany >= "11.0.0"
  3. ocaml < "5.0"
  4. minicli
  5. dune >= "2.8"
  6. dolog >= "4.0.0"
  7. cpm >= "6.0.0"
  8. batteries >= "3.2.0"

Dev Dependencies

None

Used by

None

Conflicts

None

OCaml

Innovation. Community. Security.