Posts by CodeCannon

Thumbs up to Jamie for his undying efforts on this project. Great stuff!

Hi,

Where can we download the latest MMFSDK?

In this case I'm looking for <MMF2SDK251.zip>.

Would the test pixel condition from the built in frame object work for this?

Is the object perhaps being set to x=320 more than once in a row, with its velocity gradually decreasing each time?

Try to reproduce the problem in a new mfa. You might fix it in the process. If not, upload that new mfa with the isolated problem into the file archive.

How does your object move? Do you set the position manually in the events, or is it using a built in movement?

How wide is the frame?
How wide is the window?
What is the starting position of the object?
What is the ending position of the object?

> I have a lot of stuff I need on this computer that isn't going to work on newer computers!
Have you tried a clean installation of Windows? Maybe swap out your current hard drive with another one and try it out. The old stuff might be interfering.

Hello Thrantus!

Have you tried to read/write a global value to an ini file one by one?

If your levels are arranged one after the other, you can have just one event to jump to the correct frame: Jump to frame number <Global Value I> + <Offset>. Offset could perhaps be the frame index of your first level.

What is your game called?

I think it was called the "Last Known Pursuit" method.

bigredron: Very little progress, since the last release.

The Random Pool object could be used too. Fill it with the numbers 0-99. Then for each picture, set an alterable value to a random number from the pool divided by 2 (integer division). After that you can load each picture according to its alterable value (0.png-49.png). Or set its frame index to its alterable value.

Layer > Transparency > Color to Alpha

Algul: A tutorial is in the works. There is also a project page on TDC - it has a little more info:
Please login to see this link.

Cel: OpenSteer would be more suited to a movement extension, no?

By the way, thanks to those who tested the extension examples and reported the results! It looks like the extension is stable (or at least the examples are).

Thank you for your nice comments!

Quote from MechaBowser

sounds like an amazing concept, but I can't get the AI to balance the pole correctly in the example, even after 20 generations and a lot of refreshes.

Yeah, it won't always learn the solution, which is quite a drawback, especially on very complex problems. The agents (brains) start off with random weights (neuron connections), and are tested on the problem. Then the genetic algorithm (or GA) chooses new weights for each agent, using the best ones from the last generation and mutating them a little to see whether it can improve on the last generation.

These improvements are usually very small, so the GA sometimes gets stuck on a local minima (a weak solution with nothing better nearby enough for the GA to find it). This is what "Rote escape" is for (you can see it in the properties). It is basically a separate set of mutation options that are used when the agents stop improving. You can set it up to mutate the agents to a large degree to try find better solutions that are farther away.

You can visualize this problem as a terrain with many hills and valleys. The highest hill is the best solution. The other hills are weaker solutions. The GA finds a good (but not always the best) solution by climbing up all the time. Once it gets to the top of a shallow hill it seldom leaves that hill.

Rote escape is not implemented in 0.2.1 A. I am also not sure how effective it will be, but it was described in a white paper in good light.

Hello all,

There is a small artificial intelligence extension being developed for MMF2, cryptically called "Static ANNs w/ Genetic Algorithm - DAIC" (Static-Genetic-DAIC). It's meant to be the first of a group of extensions called the "DringleTopia Artificial Intelligence Collection" (DAIC).

Would you like to try it out?

Static-Genetic-DAIC is currently in its alpha phase. So if you want to, please test it out to the limit, and report any misbehavior here. Your suggestions, bug fixes (it's open source) and feature requests are welcome too!

Don't be afraid to ask how it works - your questions will help to better the documentation. Try to be specific, though. Also, if you make something with it, even if it is very small or doesn't work, please post a link to it here if you want.

The latest information, binaries and source code for DAIC will always be available here:
Please login to see this link.

Thank you!

id3lib supports tag writing. Maybe ask the author to update the extension?

Rev to Angle = Rev/8000.0*360.0

Try adding a 1px blank border/hem around your sprites, if you are having problems with artifacts around the edges.

Have you tried asking Yves? What you want to do should be possible, because an example of it is the ability to rename alterable values.

Quote from Rabagast

I wish someone can create a good AI extension if it's possible.
It's very difficult to create a good AI. Special in a platformer.
For example in an AI extension, you can set the enemy to follow the player everywhere of course. Set the speed. How smart it is or stupid, etc.

As far as I know, there is at least one in development (besides MLP). I agree, it is hard, but maybe an AI extension with those options would be tied too closely to a specific movement or game genre? It may not be practical to make an extension like that.

When you pause a blank app with the debugger, CPU usage rises from 1% to 50%. Is this supposed to happen?

Good stuff.
Are you going to make an example game with it?