[REVIEW] Symbolic Regression/Classification C/C++ #3638

vimarsh6739 · 2021-03-19T08:38:08Z

This PR contains the implementation of the core algorithms of gplearn(tournaments + mutations + program evaluations) in cuml.
Tagging all involved: @teju85 @venkywonka @vinaydes

The goal is to complete the following tasks:

Implement program execution and metric evaluation for a given dataset on the GPU
Implement a batched version of the above for all programs in a generation
Run tournaments for program selection on the GPU
Perform all mutations on the CPU
Fit, Predict and Transform functions for api
Tests for all individual functions
Add an example demonstrating how to perform symbolic regression (a similar approach can be taken for transformation too)

GPUtester · 2021-03-19T08:38:10Z

Can one of the admins verify this patch?

…b.com/vimarsh6739/cuml into fea-ext-genetic-programming-internals(reverse pull)

venkywonka · 2021-10-28T13:17:35Z

rerun tests

venkywonka · 2021-10-28T14:00:24Z

rerun tests

teju85 · 2021-10-29T05:30:35Z

rerun tests

teju85 · 2021-10-29T12:42:18Z

@dantegd any ideas why the CI is still failing here?

tfeher

Hi @venkywonka, thanks for fixing the tests. I have one more suggestion: not to wrap device_uvectors into unique_ptr.

tfeher · 2021-10-29T13:49:36Z

cpp/test/sg/genetic/evolution_test.cu

+  std::vector<float> h_trainwts;
+  std::vector<float> h_testwts;
+
+  std::unique_ptr<rmm::device_uvector<float>> d_train, d_trainlab, d_test, d_testlab, d_trainwts,


Using unique_ptr is redundant. rmm::device_uvector already manages the lifetime of the underlying data. If the reason for wrapping it into unique_ptr, is that you do not know the size / do not have a stream at construction time, then I would simple suggest to construct it with 0 size on stream 0, and resize it later.

Wrapping it around wasn't my first choice either tamas, but since rmm::device_uvector does not have a default constructor, but the gtest struct required one, the build was failing with a "default constructor cannot be referenced -- it is a deleted function" error. Initializing it as rmm::device_uvector<float> d_train(0,0); and resize() -ing it later didn't help either. So I found a way as per here that wrapped it with unique_ptr that seemed to fix the build errors. Is the rmm::device_uvector<float> d_train(0, 0); what you meant?

If you just read the comment below what you have cited, you will see that even in that case it was possible to get rid of the extra unique_ptr wrappers. This is how it looks: link. And this is how it would look for the current PR:

diff --git a/cpp/test/sg/genetic/evolution_test.cu b/cpp/test/sg/genetic/evolution_test.cu index d131c185a..b48b0f4e2 100644 --- a/cpp/test/sg/genetic/evolution_test.cu +++ b/cpp/test/sg/genetic/evolution_test.cu @@ -39,7 +39,7 @@ namespace genetic { */ class GeneticEvolutionTest : public ::testing::Test { public: - void SetUp() override + GeneticEvolutionTest() : d_train(0, stream) { ML::Logger::get().setLevel(CUML_LEVEL_INFO); CUDA_CHECK(cudaStreamCreate(&stream)); @@ -62,7 +62,7 @@ class GeneticEvolutionTest : public ::testing::Test { h_testwts.resize(n_tst_rows, 1.0f); // Initialize device memory - d_train = std::make_unique<rmm::device_uvector<float>>(n_cols * n_tr_rows, stream); + d_train.resize(n_cols * n_tr_rows, stream); d_trainlab = std::make_unique<rmm::device_uvector<float>>(n_tr_rows, stream); d_test = std::make_unique<rmm::device_uvector<float>>(n_cols * n_tst_rows, stream); d_testlab = std::make_unique<rmm::device_uvector<float>>(n_tst_rows, stream); @@ -70,7 +70,7 @@ class GeneticEvolutionTest : public ::testing::Test { d_testwts = std::make_unique<rmm::device_uvector<float>>(n_tst_rows, stream); // Memcpy HtoD - CUDA_CHECK(cudaMemcpyAsync(d_train->data(), + CUDA_CHECK(cudaMemcpyAsync(d_train.data(), h_train.data(), n_cols * n_tr_rows * sizeof(float), cudaMemcpyHostToDevice, @@ -105,7 +105,7 @@ class GeneticEvolutionTest : public ::testing::Test { void TearDown() override { CUDA_CHECK(cudaStreamDestroy(stream)); } raft::handle_t handle; - cudaStream_t stream; + cudaStream_t stream = 0; param hyper_params; // Some mini-dataset constants @@ -244,8 +244,8 @@ class GeneticEvolutionTest : public ::testing::Test { std::vector<float> h_trainwts; std::vector<float> h_testwts; - std::unique_ptr<rmm::device_uvector<float>> d_train, d_trainlab, d_test, d_testlab, d_trainwts, - d_testwts; + rmm::device_uvector<float> d_train; + std::unique_ptr<rmm::device_uvector<float>> d_trainlab, d_test, d_testlab, d_trainwts, d_testwts; }; TEST_F(GeneticEvolutionTest, SymReg) @@ -264,7 +264,7 @@ TEST_F(GeneticEvolutionTest, SymReg) cudaEventRecord(start, stream); symFit(handle, - d_train->data(), + d_train.data(), d_trainlab->data(), d_trainwts->data(), n_tr_rows,

Okay, I got what you meant. Thanks tamas!

ah, I had just implemented what you meant in the latest commit tamas (jinx!)

tfeher · 2021-10-29T13:51:30Z

cpp/test/sg/genetic/program_test.cu

+  node* d_nodes1;
+  node* d_nodes2;
+  program_t d_progs;
+  std::unique_ptr<rmm::device_uvector<float>> d_data, d_y, d_lYpred, d_lY, d_lunitW, d_lW, dx2, dy2,


Same as above, you do not need to wrap this in unique_ptr.

dantegd · 2021-10-29T14:13:50Z

@teju85 not sure, haven't seen a timeout since increasing CI resources and a few ivfpq test modifications, let me rerun tests and if it's persistent we can push a fix

dantegd · 2021-10-29T14:13:54Z

rerun tests

venkywonka · 2021-11-01T12:26:50Z

rerun tests

tfeher

Thanks Venkat for the update! It looks good to me.

Note: I have only reviewed the C++ test.

venkywonka · 2021-11-01T17:00:52Z

rerun tests

dantegd · 2021-11-01T21:47:13Z

rerun tests

dantegd · 2021-11-02T01:45:06Z

CI is timing out on the doxygen docs check, looking into it

venkywonka · 2021-11-15T06:04:39Z

rerun tests

codecov-commenter · 2021-11-15T15:41:22Z

Codecov Report

❗ No coverage uploaded for pull request base (branch-21.12@b5f119e). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             branch-21.12    #3638   +/-   ##
===============================================
  Coverage                ?   86.03%           
===============================================
  Files                   ?      231           
  Lines                   ?    18751           
  Branches                ?        0           
===============================================
  Hits                    ?    16132           
  Misses                  ?     2619           
  Partials                ?        0

Flag	Coverage Δ
dask	`47.02% <0.00%> (?)`
non-dask	`78.74% <0.00%> (?)`

Flags with carried forward coverage won't be shown. Click here to find out more.

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b5f119e...0ca30f1. Read the comment docs.

dantegd · 2021-11-15T15:44:29Z

@gpucibot merge

@teju85

This PR contains the implementation of the core algorithms of gplearn(tournaments + mutations + program evaluations) in cuml. Tagging all involved: @teju85 @venkywonka @vinaydes The goal is to complete the following tasks: - [x] Implement program execution and metric evaluation for a given dataset on the GPU - [x] Implement a batched version of the above for all programs in a generation - [x] Run tournaments for program selection on the GPU - [x] Perform all mutations on the CPU - [x] Fit, Predict and Transform functions for api - [x] Tests for all individual functions - [x] Add an example demonstrating how to perform symbolic regression (a similar approach can be taken for transformation too) Authors: - Vimarsh Sathia (https://github.com/vimarsh6739) - Venkat (https://github.com/venkywonka) Approvers: - Robert Maynard (https://github.com/robertmaynard) - Venkat (https://github.com/venkywonka) - Thejaswi. N. S (https://github.com/teju85) - Corey J. Nolet (https://github.com/cjnolet) - Tamas Bela Feher (https://github.com/tfeher) - Dante Gama Dessavre (https://github.com/dantegd) URL: rapidsai#3638

vimarsh6739 added 3 commits March 16, 2021 02:46

Added fitness functions

0e9d987

Added single program execution kernel

2bfdcbd

Added batched tournament kernel impl

e7ce249

github-actions bot added CMake CUDA/C++ labels Mar 19, 2021

Merge branch 'branch-0.19' into fea-ext-genetic-programming-internals

b8fa564

vimarsh6739 added 6 commits April 2, 2021 03:09

Added point mutations and subtree extraction

4d55122

Removed some compilation bugs

befe036

Added crossover

07bb099

Merge branch 'fea-ext-genetic-programming-internals' of https://githu…

82ef099

…b.com/vimarsh6739/cuml into fea-ext-genetic-programming-internals(reverse pull)

Added hoist mutations

e54c0bd

Build random full depth programs

e7cc561

teju85 changed the base branch from branch-0.19 to branch-0.20 April 6, 2021 02:02

teju85 added New Algorithm For tracking new algorithms that will be added to our existing collection non-breaking Non-breaking change Experimental Used to denote experimental features labels Apr 6, 2021

vimarsh6739 added 5 commits April 10, 2021 01:12

Pass twister directly when mutating

28fdbba

Mutation decision before tournaments

764abaf

Host uses only call by references

1e9b0ae

Batched execution done

800811b

Batched loss function impl left

799ee31

vimarsh6739 changed the title ~~WIP: Adding genetic programming internal cpp code~~ [WIP] Adding genetic programming internal cpp code Apr 12, 2021

vimarsh6739 added 6 commits April 13, 2021 00:20

Added batched version of all loss functions

cd66f94

Initial generation covered

62830de

Testing template + double free bugfix in program

b71f96c

Added tests for loss functions

01c4804

Fixed row broadcasting bug

9ae8007

Increased tolerance to 5%. To optimize log loss

c5301d1

venkywonka added 3 commits October 26, 2021 04:57

Merge branch 'branch-21.12' into fea-ext-genetic-programming-internals

3b55e2f

fix memleak and change all device allocation to rmm

8e96272

fix memory leak of last generation outside in test

b756c13

venkywonka force-pushed the fea-ext-genetic-programming-internals branch from aba3cda to b756c13 Compare October 28, 2021 11:13

add a doxygen note detailing the memory allocation behaviour

a875351

tfeher requested changes Oct 29, 2021

View reviewed changes

remove unique_ptr

57e88f9

tfeher approved these changes Nov 1, 2021

View reviewed changes

venkywonka added 2 commits November 15, 2021 15:00

Merge branch 'branch-21.12' into fea-ext-genetic-programming-internals

64837aa

accounting for raft updates on matrix,stats and random

0ca30f1

dantegd approved these changes Nov 15, 2021

View reviewed changes

rapids-bot bot merged commit 80cb593 into rapidsai:branch-21.12 Nov 15, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[REVIEW] Symbolic Regression/Classification C/C++ #3638

[REVIEW] Symbolic Regression/Classification C/C++ #3638

vimarsh6739 commented Mar 19, 2021 •

edited

Loading

GPUtester commented Mar 19, 2021

venkywonka commented Oct 28, 2021

venkywonka commented Oct 28, 2021

teju85 commented Oct 29, 2021

teju85 commented Oct 29, 2021

tfeher left a comment

tfeher Oct 29, 2021

venkywonka Oct 29, 2021 •

edited

Loading

tfeher Nov 1, 2021

venkywonka Nov 1, 2021 •

edited

Loading

venkywonka Nov 1, 2021

tfeher Oct 29, 2021

dantegd commented Oct 29, 2021

dantegd commented Oct 29, 2021

venkywonka commented Nov 1, 2021

tfeher left a comment •

edited

Loading

venkywonka commented Nov 1, 2021

dantegd commented Nov 1, 2021

dantegd commented Nov 2, 2021

venkywonka commented Nov 15, 2021

codecov-commenter commented Nov 15, 2021

dantegd commented Nov 15, 2021

[REVIEW] Symbolic Regression/Classification C/C++ #3638

[REVIEW] Symbolic Regression/Classification C/C++ #3638

Conversation

vimarsh6739 commented Mar 19, 2021 • edited Loading

GPUtester commented Mar 19, 2021

venkywonka commented Oct 28, 2021

venkywonka commented Oct 28, 2021

teju85 commented Oct 29, 2021

teju85 commented Oct 29, 2021

tfeher left a comment

Choose a reason for hiding this comment

tfeher Oct 29, 2021

Choose a reason for hiding this comment

venkywonka Oct 29, 2021 • edited Loading

Choose a reason for hiding this comment

tfeher Nov 1, 2021

Choose a reason for hiding this comment

venkywonka Nov 1, 2021 • edited Loading

Choose a reason for hiding this comment

venkywonka Nov 1, 2021

Choose a reason for hiding this comment

tfeher Oct 29, 2021

Choose a reason for hiding this comment

dantegd commented Oct 29, 2021

dantegd commented Oct 29, 2021

venkywonka commented Nov 1, 2021

tfeher left a comment • edited Loading

Choose a reason for hiding this comment

venkywonka commented Nov 1, 2021

dantegd commented Nov 1, 2021

dantegd commented Nov 2, 2021

venkywonka commented Nov 15, 2021

codecov-commenter commented Nov 15, 2021

Codecov Report

dantegd commented Nov 15, 2021

vimarsh6739 commented Mar 19, 2021 •

edited

Loading

venkywonka Oct 29, 2021 •

edited

Loading

venkywonka Nov 1, 2021 •

edited

Loading

tfeher left a comment •

edited

Loading