Chess programs or other programs with capable heruristics for calculators

[pier4r]

I have one ti89, that surely is quite capable but first I want to wear off other calcs with programming and math; thus I try to use it as I can in the meantime. Thanks to the ti89 community, the calc has a lot of applications through which one can use the device.
One that let me use the calc from time to time is tichess 4.17e http://tict.ticalc.org/projects.html#ref_latest . Could be that the program is not the best chess entity that could be programmed for the ti89 but most likely is the best chess entity actually implemented on it.

What about other calculators similar (or more capable) to the ti89? Of course all the calc compatible with the ti89 can run lichess as well.

For the 50g (and compatible) I found mostly one program with a serious heuristic: MLchess https://www.hpcalc.org/details/3067
For the prime I got none.
What about the casio calcs? I didn't find any clear information of programs there.

Moreover I was also interested in programs having capable heristics in a search domain that is not trivial (thus other games/problems). Do you have any known example? (if possible with available source code)

[Valentin Albillo]

(10-14-2020 09:14 PM)pier4r Wrote:  Moreover I was also interested in programs having capable heristics in a search domain that is not trivial (thus other games/problems). Do you have any known example? (if possible with available source code)

Just a clarification: what is your definition of "heuristics" ? Do you consider the Alpha–beta pruning within a minimax search in chess to be a heuristic or an algorithm ? Do you consider programs that solve, say, Sudokus using bitboard-handling and recursion, to be using heuristics or algorithms ?

Also, are you interested in source code for any kind of calc and any brands ? That's too wide a mark.

Regards.
V.

[pier4r]

Well for heuristic I mean something that is not going through every possibility (or the majority of them) of a large search space (I know it is vague, but I cannot set hard limits). Moreover since we are talking about calculators with limited ram and computing power (maybe aside from later models), the heuristic has to be limited otherwise it can hardly get deep in the search space.
Of course there are also well known formulas, like finding roots with the newton method. That is also an heuristic but it is pretty effective and well known.

I know that it could be a wide search, but let's start at first and let's see how it goes. One thing leads to another. And yes I am interested in whatever calc model and non-small problem . Starting with chess maybe.

[Valentin Albillo]

(10-15-2020 07:01 AM)pier4r Wrote:  Well for heuristic I mean something that is not going through every possibility (or the majority of them) of a large search space

Understood. That perfectly describes chess and in general games where the move tree is impractically big to go for a brute-force approach. Among many other subject matters wholly unrelated to games, of course

Quote:Of course there are also well known formulas, like finding roots with the newton method. That is also an heuristic [...]

Nope. Newton's method is an algorithm, not some heuristic.

Quote:I am interested in whatever calc model and non-small problem . Starting with chess maybe.

As I said above, any game having a big move tree will do: board games like chess, checkers, go, Reversi (aka Othello), dominoes-related games, many card games, etc. And as for non-games, any program which deals with problems like the Travelling salesman problem and other such would also qualify.

Good luck and have a nice weekend. Take care.
V.

[pier4r]

(10-15-2020 10:21 PM)Valentin Albillo Wrote:  Nope. Newton's method is an algorithm, not some heuristic.

I am not sure I understand. An heuristic is always an algorithm on a computer (or even on paper, if one wants to follow it). What is the difference?

Aside from the precision of course. An algorithm can be a closed formula that is very precise, but an algorithm can be also an implementation fo an heuristic, for what I know.

[toml_12953]

(10-16-2020 08:09 AM)pier4r Wrote:  

(10-15-2020 10:21 PM)Valentin Albillo Wrote:  Nope. Newton's method is an algorithm, not some heuristic.

I am not sure I understand. An heuristic is always an algorithm on a computer (or even on paper, if one wants to follow it). What is the difference?

Although a heuristic is an algorithm, a lot of people think of a heuristic as a proper subset of algorithms. All heuristics are algorithms but not all algorithms are heuristics. They differentiate algorithm from heuristic in that an algorithm is static; you write the solution to a problem once and it solves the problem the same way every time. A heuristic is self-modifying; it can learn by repeated trial and error so a heuristic chess program would learn by the outcome of every game and get better and better. The goal would be to have the program become unbeatable after many games, I would assume.

[pier4r]

(10-16-2020 11:53 AM)toml_12953 Wrote:  Although a heuristic is an algorithm, a lot of people think of a heuristic as a proper subset of algorithms. All heuristics are algorithms but not all algorithms are heuristics.

Ok, I can see this somehow.

Quote:They differentiate algorithm from heuristic in that an algorithm is static; you write the solution to a problem once and it solves the problem the same way every time. A heuristic is self-modifying; it can learn by repeated trial and error so a heuristic chess program would learn by the outcome of every game and get better and better. The goal would be to have the program become unbeatable after many games, I would assume.

I am not convinced by this. Picking chess as example, a lot of evaluation functions and others were static, that is: given the same input, they will give out always the same output, without improving. This setting aside possible non-determinism if one runs multithread applications.

Those were also called heuristics. Thus on this point I could agree with machine learning, but all other approaches would be then "only" algorithms ?