kdeldycke / chessboard

:game_die: CLI to solve combinatoric chess puzzles.

My name is Kevin Deldycke, I'm a high-tech Engineering Manager.

Short Bio

I've held various positions in several industries and sectors as VP, Product Manager, Founding Engineer and Consultant.

After 15+ years of relevant engineering work, and 3+ years of hands-on management experience, I now build teams from scratch with self-sustaining culture, and my intent is to keep doing so.

Enabling engineers, leading managers, and make them feel productive is the most impactful thing I can bring to the table. So I'm always on the lookout for a high-tech management position, at a fast growing start-up with ambitious vision.

My most relevant experience was in growing a team from 0 to 12 at a world-class cloud computing provider. We built up the whole IAM ecosystem, as well as the Billing and Payment stack.

Achievements

• 💰 Designed, implemented and operated Scaleway's billing pipeline, generating multi-million dollars of revenue (and growing!). The whole thing. All by myself.

• 🚀 Then I built a team of 12 people to enhance and expand on the full Billing, Payment and IAM domain of the company. In doing so, I was given the fancy title of Product Manager, then VP.

• 🤝 Established lasting trust and loyalty. No one from my team resigned under my authority, in a company with a very high employee turnover rate.

• 🥇 World trending GitHub developer, and top-1% open-source developer.

• 🌟 21,000+ GitHub stars. The most popular repository being the Falsehoods Programmers Believe in. It is regularly featured on popular media.

• 💵 Wrote the best open-source payroll software in the world.

• 🌐 Created and managed the African subsidiary of my first employer.

• 🛰 Authored code that's running on Mars.

• 🤡 Wasted 6 years solving an XKCD joke.

More

• Interviews
• LinkedIn resume

chessboard's Issues

Add support for PyPy

This is a big feature that is unlikely to be addressed soon, as Numpy doesn't support PyPy yet.

Still, work seems in progress. See:

• http://buildbot.pypy.org/numpy-status/latest.html
• https://bitbucket.org/pypy/numpy
• https://github.com/rvianello/conda-pypy

Use Appveyor CI to test Windows targets

See: https://github.com/rmcgibbo/python-appveyor-conda-example

Start with pieces of biggest territory coverage

Starting the tree exploration with pieces having higher degree of territory coverage might help discarding branches faster.

Try this approach and measure it.

Needs #3 to be addressed first.

Plug-in logger facilities

To help debug, add proper logging facilities from the Python module of the same name. Will provide extended configurability.

Auto-deploy tagged released to PyPi

Looks like there is a way to auto-deploy releases to PyPi, as per:

• https://github.com/jaraco/skeleton/blob/746dd7999f9db23276144ee2160920bd01ed860c/.travis.yml#L13-L23
• https://docs.travis-ci.com/user/deployment/pypi/

Exploit symetries

All pieces are moveable in the same fashion in the 4 quadrants. Exploit vertical and horizontal symmetries to reduce the search space.

A dynamic programming solution to the n-queens problem

http://www.cs.cornell.edu/~rdz/Papers/RZ-IPL92.pdf
https://news.ycombinator.com/item?id=13276975

Add Python 3 compatibility

Module is Python 2 only for the moment:

• https://github.com/kdeldycke/chessboard/blob/v0.8.0/.travis.yml#L2-L3
• https://github.com/kdeldycke/chessboard/blob/v0.8.0/setup.py#L75-L77

Cut-short rotten branches

The current tree() iterator produce a cartesian product. It should instead explore branches of the tree and abandon them as soon as one doesn't fit our criterion.

Use lighter datastructure

For the sake of simplicity and readability, I used natural types like lists of strings and booleans to materialize intermediate data.

To speed things up, replace these data structures with lower-level ones like bytearray and such.

Explore use of space filling curves

See:

• https://en.wikipedia.org/wiki/Space-filling_curve
• https://en.wikipedia.org/wiki/Hilbert_curve
• https://en.wikipedia.org/wiki/Z-order_curve

Use boltons.ecoutils

...to gather software and hardware metadata in benchmark.

See: https://boltons.readthedocs.io/en/latest/ecoutils.html#profile-generation

Distribute load on available cores

Current solver only works on a single core. Leverage all cores from the CPU.

Deduplicate permutations of same-type pieces

Group permutations are pre-computed by the grouped_permutations() method, introduced in commit 7fa785a. Some permutations produced are already invalid. We can prune those from our sets before running the brute force cartesian product tree() method.

Binary Optimization

See:

• https://www.chessprogramming.org/Magic_Bitboards
• https://news.ycombinator.com/item?id=16860317

Broken build on Python 3.3

Build on Python 3.3 is broken: https://travis-ci.org/kdeldycke/chessboard/jobs/137079508

That's because of the sphinx >= 1.4 requirement added in conda.yml in 8dd2029 . Unfortunately, the package seems to be build for the Python 3.3 distribution by anaconda: https://anaconda.org/anaconda/sphinx/files