Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 24, 2020

• Update README.md

  benjaminpatrickevans

  authored
  61ed340

  Copy full SHA for 61ed340

Commits on Oct 22, 2019

• Stop bayesian optimisation if not progressing after x iterations. Also exit cleanly (with best parameters) if we time out when optimising

  benjaminpatrickevans

  committed
  5016c8c

  Copy full SHA for 5016c8c

• Fix case when only 1 generation is run and fitnesses are invalid

  benjaminpatrickevans

  committed
  cc07b15

  Copy full SHA for cc07b15

Commits on Oct 21, 2019

• Novelty search.

  benjaminpatrickevans

  committed
  0896c3a

  Copy full SHA for 0896c3a

Commits on Oct 20, 2019

• Remove caching since it doesnt make sense for adaptive fitness. Cleanup timing out process

  benjaminpatrickevans

  committed
  3f06c5b

  Copy full SHA for 3f06c5b

Commits on Oct 19, 2019

• Cleanup complexity measure calculation. New random seed each time performing bayes

  benjaminpatrickevans

  committed
  f3fba5f

  Copy full SHA for f3fba5f

• Removed the idea of crossover, refactored the search and base class to reflect this. Also added a new seach based on googles idea in 'Large scale evolution of image classifiers'.

  benjaminpatrickevans

  committed
  9e8d13b

  Copy full SHA for 9e8d13b

Commits on Oct 18, 2019

• Refactored bayesian optimisation to own class

  benjaminpatrickevans

  committed
  1a2eefa

  Copy full SHA for 1a2eefa

• Bayesian optimisation as a mutation operator

  benjaminpatrickevans

  committed
  dd30cb4

  Copy full SHA for dd30cb4

• Run bayesian search on the best resulting individual after evolution

  benjaminpatrickevans

  committed
  000fdc5

  Copy full SHA for 000fdc5

• Choose appropriatew bayesian sampler based on value type we are trying to optimise:

  benjaminpatrickevans

  committed
  7f875f9

  Copy full SHA for 7f875f9

• Use the values found by GP as the starting point for bayesian optimisation

  benjaminpatrickevans

  committed
  5699b12

  Copy full SHA for 5699b12

• Initial work on adding bayesian optimisation to hyperparameters

  benjaminpatrickevans

  committed
  5cf4486

  Copy full SHA for 5cf4486

Commits on Oct 15, 2019

• New crossover which just mutates the best of the 2 parents

  benjaminpatrickevans

  committed
  1f8e37b

  Copy full SHA for 1f8e37b

• Select mutation randomly from list of mutation types

  benjaminpatrickevans

  committed
  adb73f1

  Copy full SHA for adb73f1

• Added mutation which can grow an ensemble (mutInsert)

  benjaminpatrickevans

  committed
  3a35f00

  Copy full SHA for 3a35f00

• Try the various mutation operators until we get a unique one. Renamed deapfix to customdeap

  benjaminpatrickevans

  committed
  3408a85

  Copy full SHA for 3408a85

• Added mutation which can shrink combiner nodes to one of their children

  benjaminpatrickevans

  committed
  0f2fabc

  Copy full SHA for 0f2fabc

• Fixed bug with getting children indices in mutNodeReplacement

  benjaminpatrickevans

  committed
  7214bcf

  Copy full SHA for 7214bcf

• Added a mutation operator which replaces a node but keeps the children

  benjaminpatrickevans

  committed
  b77dfa1

  Copy full SHA for b77dfa1

• Visualisation of final model added

  benjaminpatrickevans

  committed
  e4de3cc

  Copy full SHA for e4de3cc

Commits on Oct 14, 2019

• Add a max eval time for individual models. Disable PoylnomialFeatures in the case of there being >32 features

  benjaminpatrickevans

  committed
  b9d6cbb

  Copy full SHA for b9d6cbb

Commits on Oct 13, 2019

• Added Stacking with LR. Added PolynomialFeatures.

  benjaminpatrickevans

  committed
  ab67968

  Copy full SHA for ab67968

• Try eval() in mutate several times for each individual until unique one generated

  benjaminpatrickevans

  committed
  f336bf4

  Copy full SHA for f336bf4

• Added diversity crossover and mutation, which will try the operators repeatedly until they can generate individuals which have not yet existed

  benjaminpatrickevans

  committed
  35fbceb

  Copy full SHA for 35fbceb

Commits on Oct 12, 2019

• Ensure we are always selecting atleast 1 feature when performing feature selection

  benjaminpatrickevans

  committed
  9ae2b18

  Copy full SHA for 9ae2b18

• Added data preprocessor.

  benjaminpatrickevans

  committed
  2e549e9

  Copy full SHA for 2e549e9

• Refactoring so adding estimators and selectors is a similar process. Began extending for data preprocessors

  benjaminpatrickevans

  committed
  152fb50

  Copy full SHA for 152fb50

Commits on Oct 11, 2019

• Added feature selection

  benjaminpatrickevans

  committed
  9516caf

  Copy full SHA for 9516caf

• Added dummy and random feature selection steps which occur before classificiation

  benjaminpatrickevans

  committed
  ae7f91f

  Copy full SHA for ae7f91f

• Refactored combiners. Began introducing pipelines

  benjaminpatrickevans

  committed
  b05672c

  Copy full SHA for b05672c

Commits on Oct 10, 2019

• Expanded possible classifiers

  benjaminpatrickevans

  committed
  2072058

  Copy full SHA for 2072058

Commits on Oct 9, 2019

• Option to use original features or just predictions in stacking

  benjaminpatrickevans

  committed
  13e1108

  Copy full SHA for 13e1108

• Added Stacking

  benjaminpatrickevans

  committed
  ef30875

  Copy full SHA for ef30875

Commits on Oct 8, 2019

• Regression added.

  benjaminpatrickevans

  committed
  0bbc6f8

  Copy full SHA for 0bbc6f8

Pagination

Previous

Next