Hyperparameter tuning for custom strategies - #329

Merged

benvanwerkhoven merged 290 commits into

hyperparametertuning_custom_strategies

Sep 3, 2025

This pull request is big! We're only showing the most recent 250 commits

Commits on Nov 6, 2024

Setup structure for Tensorspace in Searchspace
fjwillemsen
committed
009cf01

Commits on Nov 7, 2024

Commits on Nov 8, 2024

Completed implementation of mixed-type handling and handling of invalid and evaluated configurations
fjwillemsen
committed
196af62
Added docstrings, improved formatting
fjwillemsen
committed
55a5c1a
Extended strategies test to test for ability to handle non-numeric and mixed parameter values
fjwillemsen
committed
d64f783
Mixed-type parameters are not converted to numeric constraints
fjwillemsen
committed
e95ab30
CostFunc can now encode and decode non-numeric configurations for strategies that require only numerics
fjwillemsen
committed
10a6a5c
Fixed logging statements, improved formatting
fjwillemsen
committed
6ae3ba6
Improved the performance of get_bounds
fjwillemsen
committed
4873a20
Applied non-numeric encoding in differential evolution to handle non-numeric parameter values
fjwillemsen
committed
bae7e96
Implemented automatic conversion to multiple types for encoded tensor parameter lookup
fjwillemsen
committed
7eb7ef7
Added tests for Searchspace tensor encoding and conversion
fjwillemsen
committed
91d3ce4
Seperated strategies and runners test cache file
fjwillemsen
committed
80d514e
Implemented handling of categorical parameters
fjwillemsen
committed
a489252
Implemented variational GP and likelihood
fjwillemsen
committed
68aee14
Using LogExpectedImprovement to avoid stability issues
fjwillemsen
committed
b9c012d

Commits on Nov 9, 2024

Implemented tensor space bounds in searchspace
fjwillemsen
committed
41ce663
Implemented normalization for input features
fjwillemsen
committed
07ef1d4
Tensorspace is reduced by removing inconsequential parameters
fjwillemsen
committed
721d072
Extended strategies tests to include single parameter value
fjwillemsen
committed
2434b3b
Fixed an indexing error for tensorspace bounds
fjwillemsen
committed
1679751
Extended searchspace tests to include single parameter value
fjwillemsen
committed
2b816a6

Commits on Nov 10, 2024

Commits on Nov 12, 2024

Commits on Nov 13, 2024

Commits on Nov 19, 2024

Commits on Nov 22, 2024

Added BO transfer learning strategy
fjwillemsen
committed
1162ece
Implemented optionally constructing a searchspace from a cache dictionary
fjwillemsen
committed
62fa135
Implemented construction of Searchspaces from caches
fjwillemsen
committed
57a262f
Transfer learning inputs and outcomes are represented in Tensors
fjwillemsen
committed
964a6ee
More general approach to model and likelihood initialization to make it suitable for transfer learning
fjwillemsen
committed
24c6767
Fitting a model for each base transfer learning task
fjwillemsen
committed
dc4b4c7
Account for invalid configurations in base task caches
fjwillemsen
committed
e21a605
Implement main RGPE BO loop
fjwillemsen
committed
e3cfe91
Improved the efficiency of taking initial sample
fjwillemsen
committed
2334214
Use of state dictionary is made optional
fjwillemsen
committed
c78a18c
Renamed RGPE strategy
fjwillemsen
committed
8416098
Implemented new transfer learning strategy with multiple independent GPs
fjwillemsen
committed
dc000b7

Commits on Nov 23, 2024

Removed redundant min/max results adjustment
fjwillemsen
committed
aa30ec2

Commits on Nov 26, 2024

Commits on Mar 5, 2025

BO update
fjwillemsen
committed
a08953e
Improved conversion of tunable parameter
fjwillemsen
committed
ecd7802

Commits on Mar 6, 2025

Extended and improved conversion of T1 arguments, improved error reporting on problem size differences
fjwillemsen
committed
b7cda36

Commits on Mar 7, 2025

Improved selection of transfer learning caches
fjwillemsen
committed
8836ce2
Fixed an error with quotes in an f-string
fjwillemsen
committed
539aed3
Fixed torch import error due to Tensor type hint
fjwillemsen
committed
435b56b
Fixed torch import error due to Tensor type hint
fjwillemsen
committed
3c48b49
Fixed torch import error due to Tensor type hint
fjwillemsen
committed
388f325
Fixed torch import error due to Tensor type hint
fjwillemsen
committed
373782f
Merge with searchspace_experiments
fjwillemsen
committed
874c390
Merge with searchspace_experiments
fjwillemsen
committed
db3abb3
Loosened required positional arguments
fjwillemsen
committed
c692ba6
Changed benchmarks location for hypertuner
fjwillemsen
committed
fe113e6
Used hip-python-fork package as hip-python is not available
fjwillemsen
committed
5e65abd
Removed transfer learning references
fjwillemsen
committed
0ba00a0
add support for user-defined optimization algorithms
benvanwerkhoven
committed
1517565
Updated pyproject
fjwillemsen
committed
6633bed
Merge branch 'searchspace_experiments' into hyperparametertuning
fjwillemsen
committed
23f557f
Adjusted hyper.py for paper
fjwillemsen
committed
c39ac5a
Extended hypertuner with additional kernels, adjusted for benchmark_hub
fjwillemsen
committed
cc19515
Merge branch 'hyperparametertuning' of https://github.com/benvanwerkhoven/kernel_tuner into hyperparametertuning
fjwillemsen
committed
1433930

Commits on Mar 8, 2025

Commits on Mar 13, 2025

Updated hyperparameters
fjwillemsen
committed
1056269
Updated search spaces used in hyperparameter tuning and number of repeats
fjwillemsen
committed
7ce2234

Commits on Mar 15, 2025

Added bayes_opt to hyperparamtuning
fjwillemsen
committed
1ed1893

Commits on Mar 20, 2025

Avoid requesting more random samples than the searchspace size
fjwillemsen
committed
651c42c
Clearer message when exceeding the stop criterion
fjwillemsen
committed
b953a69
Add soft maximum function evaluations limit to dual annealing
fjwillemsen
committed
a401008
Improved rounding of encoded parameter values
fjwillemsen
committed
425b4f4
Merge of doc requirements files
fjwillemsen
committed
0b7ec15
Updated pyproject and requirements files
fjwillemsen
committed
3bba923
Improved assertion error message
fjwillemsen
committed
64dfd95
Added logging in case default block size restriction is added
fjwillemsen
committed
5a83d36
Adjusted path to benchmarking kernels
fjwillemsen
committed
5e3512b
Automatically adjust genetic algorithm popsize for smaller search spaces
fjwillemsen
committed
bff6d7b
Updated poetry configuration fields to project configuration fields, updated dependencies
fjwillemsen
committed
8ddce18
Removed not yet fully implemented bayesian optimization references, made tests work
fjwillemsen
committed
19470e4
Avoid import of whole util module
fjwillemsen
committed
d2bb76a
Avoid import of whole util module
fjwillemsen
committed
58f147f
Avoid import of whole util module
fjwillemsen
committed
a48394a

Commits on Mar 25, 2025

Commits on Mar 26, 2025

Commits on Apr 3, 2025

Merge with recent additions to searchspace_experiments
fjwillemsen
committed
26914be

Commits on Apr 30, 2025

Commits on May 1, 2025

fix test_mutate
benvanwerkhoven
committed
67a5070

Commits on May 12, 2025

constraint-aware variants for pso, firefly, and sa
benvanwerkhoven
committed
939ea19
remove unused variable
benvanwerkhoven
committed
b358265
Added objective performance keys
fjwillemsen
committed
2d24ae9

Commits on May 13, 2025

Commits on May 14, 2025

Added firefly to hyperparameter tuning, various minor improvements
fjwillemsen
committed
f6811ab

Commits on May 15, 2025

Added explicit restrictions definition to hyperparameter tuning
fjwillemsen
committed
e4af9f7

Commits on May 16, 2025

Updated tune_kernel_T1 to be more broadly applicable
fjwillemsen
committed
5f3b6fc

Commits on May 24, 2025

Updated hyperparameters to newly tuned defaults
fjwillemsen
committed
7f3a4a3

Commits on May 28, 2025

Commits on May 30, 2025

Commits on Jun 2, 2025

Commits on Jun 4, 2025

Commits on Jun 5, 2025

Improved dynamic import of modules for custom strategies
fjwillemsen
committed
e8ff6e7
Expanded custom optimizer test to T1 input format
fjwillemsen
committed
160d4b7

Commits on Jun 12, 2025

Commits on Jun 16, 2025

Commits on Jun 17, 2025

Initial version of using external package strategies
fjwillemsen
committed
268bf67

Commits on Jun 19, 2025

Implemented the pyatf_strategies, which enables using the pyATF strategies in Kernel Tuner
fjwillemsen
committed
20ea709

Commits on Jun 20, 2025

Commits on Jun 21, 2025

Implemented caching of PyATF searchspace object, both storage and retrieval
fjwillemsen
committed
934be28

Commits on Jun 23, 2025

Implemented passing whether or not to use the searchspace cache as a hyperparameter
fjwillemsen
committed
1d916bd

Commits on Jun 25, 2025

Commits on Jun 26, 2025

Commits on Jun 27, 2025

Implemented building multiple neighbor index caches, optional parameter to build full cache
fjwillemsen
committed
8954a46

Commits on Jul 3, 2025

Improvements for Numpy 2.0 compatibility
fjwillemsen
committed
e2924e0
Completed merge with master
fjwillemsen
committed
0d9b90b
replace differential evolution strategy
benvanwerkhoven
committed
c39b87e

Commits on Jul 4, 2025

formatted with black
benvanwerkhoven
committed
a81765a
fix parsing diff_evo method argument
benvanwerkhoven
committed
655fcc0
add test for parse method
benvanwerkhoven
committed
2fdee80
add support for x0 starting point
benvanwerkhoven
committed
539735c
add constraint-awareness
benvanwerkhoven
committed
baf628d
merged master into custom_diff_evo
benvanwerkhoven
committed
962e5f9
Merge with custom_diff_evo
fjwillemsen
committed
5798286
Improved bounds check to deal with non-numericals, other improvements
fjwillemsen
committed
7d3861d
Replaced undefined references to self
fjwillemsen
committed
958f8bd
LHS sampling, enforce trial population diversity, avoid getting stuck
benvanwerkhoven
committed
ea7a69d
Moved cache_filename reference to top-level for import elsewhere
fjwillemsen
committed
fd96b81
code quality improvements
benvanwerkhoven
committed
26c8127

Commits on Jul 7, 2025

Merge with latest changed from diff_evo
fjwillemsen
committed
05caa5f

Commits on Jul 10, 2025

Commits on Jul 11, 2025

Commits on Jul 17, 2025

Commits on Jul 18, 2025

Improved matching problem size check
fjwillemsen
committed
8bd5991

Commits on Jul 19, 2025

Commits on Jul 22, 2025

Implemented a fast distributed random sample technique suitable for the discrete constraint search spaces
fjwillemsen
committed
3b1aeed

Commits on Jul 25, 2025

Disabled full validate on load for hyperparameter tuning
fjwillemsen
committed
c044ef8
Extended hyperparameters for optimization algorithms in paper
fjwillemsen
committed
cde7823
Updated required python-constraint version
fjwillemsen
committed
f56c1fd

Commits on Aug 1, 2025

Set new hyperparameter values for various strategies after more extensive training
fjwillemsen
committed
2097092

Commits on Aug 21, 2025

Tweaked number of valid search iterations for hyperparameter tuning
fjwillemsen
committed
ad41cfe

Commits on Sep 3, 2025

Hyperparametertuning custom strategies (#325)
nikivanstein
authored
301c421
Deleted two unnecessary files
fjwillemsen
committed
b0de141