[RFC]: Refactor random number generation in JS benchmarks for stats/base/dists/poisson

[anandkaranubc]

Description

This RFC proposes improving random number generation in JS benchmarks for stats/base/dists/poisson.

Context: At present, in the remaining packages, random number generation in JS benchmarks occurs inside the benchmarking loop. Since random number generation is an expensive operation, it should be moved out of the benchmarking loops and initialized beforehand to avoid interfering with the results.

When adding support, the following tasks should be completed:

1. Move random number generation out of the benchmarking loops and initialize it before the benchmarks.

2. Ensure that the generated random values use the same range as the existing values to maintain consistency.

3. Use uniform and discreteUniform from @stdlib/random/base/uniform and @stdlib/random/base/discrete-uniform instead of randu expressions.

   • For example, replace ( randu() * 10.0 + EPS ) with uniform( EPS, 10.0 ).
   • Replace ceil( randu() * 10.0 + EPS ) with discreteUniform( 1, 10 ).

To provide a concrete example of what a PR implementing the desired changes should contain, see #4837 and #4955, which provide examples for using both uniform and discreteUniform.

Related Issues

#4837, #4955

Questions

No.

Other

Once the implementation is complete, you should be able to run the following make commands:

Build Native Add-on

NODE_ADDONS_PATTERN="@stdlib/stats/base/dists/poisson/*" make install-node-addons

Run JavaScript Benchmarks

make benchmark-javascript-files FILES="$(pwd)/lib/node_modules/@stdlib/stats/base/dists/poisson/*/benchmark/benchmark.js"

Run JavaScript Native Benchmarks

make benchmark-javascript-files FILES="$(pwd)/lib/node_modules/@stdlib/stats/base/dists/poisson/*/benchmark/benchmark.native.js"

Note: If running benchmarks results in an error, it is likely due to the random number generators producing values where the functions are not defined. To resolve this, check the relevant benchmark files, adjust the input value ranges to ensure they remain within the domain where the functions are valid, and then rerun the benchmarks. Reference: #4955

Checklist

• I have read and understood the Code of Conduct.
• Searched for existing issues and pull requests.
• The issue name begins with RFC:.

[gagan-bhullar-tech]

I would like to work on this issue. Please assign.

[Krishna-Sharma-g]

hello @anandkaranubc i would like to work on this so basically i have to change every benchmark like as you said for every file inside the poisson like cdf , ctor , entropy and all so i have done it for one that is cdf's benchmark can you please check that i have to do it like this or i have to do something else

like here i changed the randu with the uniform and the ciel one with the discreteUniform one and also i initialized them before the benchmarking loop

can you please share your thoughts !!

[Krishna-Sharma-g]

@kgryte can you also review it please

[anandkaranubc]

can you please share your thoughts !!

Thanks, @gautam-krishna-sharma, for looking into how to solve this issue. Here are the changes you need to make:

• The variable names will remain the same. xValues and lambdaValues should still be called x and lambda.
• Use Float64Array instead of Array for array initialization.
• The lengths of the arrays should be set using a len variable, initialized to 100, instead of b.iterations.
• You do not need to add comments explaining what is being done.
• Ensure that you follow the code conventions, such as adding extra space between the start and end of brackets, like x[ i ] = discreteUniform( 1, 100 ).

The reference issues will be a good starting point to guide your changes. Good luck!

[Krishna-Sharma-g]

okay thankyou @anandkaranubc i will surely do these changes and will look into the reference issue

[pranav-1720]

Created a PR following the examples of the other refactors.