connect4 | Solving board games like Connect4 using Deep Reinforcement | Machine Learning library

The main issue is in these lines of code:

The problem here is actually in the different function grid is returned from. You must have made some mistake there, which causes that different function return something of the form [1, 6, 3, 8, 3], whereas your play function assumes something in the form of [[1, 5, 6, 2, 10], [1, 5, 6, 2, 10], [1, 5, 6, 2, 10], [1, 5, 6, 2, 10]].

You can use animationend to check when the animation end and move the ball element between the divs

You can approximate this using an SVG filter (more detail here: https://dev.to/afif/css-shapes-with-rounded-corners-56h)

As Ouroborus pointed out, at depth 5 it sees that it loses no matter what move it plays. So now it choses the first move in your list of possible moves since all results return -1000.

If you want it to always find the longest route to lose then you need to return -1000 + depth if you lose, and 1000 - depth if you win. Then your AI will always chose the longest route to losing (and the quickest to winning if there are more than 1 way to win).

The comment from Shai is correct. You do not need to use Conv3D layer here. The shape of your Conv3D filters would violate the calculation of size after application of a convolutional filter by reducing at least 1 dimension to less than 1 which is why you are getting your error (you can't multiple with a value that does not exist).

Simply using the original model implementation should work for you.

Similar to images with 3 color bands, these are typically not processed with Conv3d (maybe a different case with hyperspectral images, but that is not relevant here). There is some discussion about how to treat each of the color bands, and you can affect this in a variety of ways.

For example, adjusting the groups argument of the Conv2D layer at instantiation will change the connections between in_channels and out_channels of the layer and which are convolved to which, as per their documentation: https://pytorch.org/docs/stable/generated/torch.nn.Conv2d.html.

You may be able to optimize this in your model, or otherwise experiment with it.

In any case, simply using the existing implementation with Conv2D should be fine for you. Conv3D is typically utilized in the case of 3 spatial dimensions, or sometimes 2 spatial and 1 temporal dimension. While your case is kind of like a limited version of 3 spacial dimensions, it is not necessarily the same as, say, a 3D vector-field of fluid flow in regards to how each "pixel" has some spatial relevance/correlation to its neighboring "pixels". Your "spatial pixels" have a somewhat different kind of relevance or correlation mapping than this.

Next time you have this kind of issue try to use a systematic approach to define where is your bug:

• Is your draw() function broken? To check that put something in your board and see if it shows. For example add board[1][1]=1 in your setup() and you'll see a circle appearing so the issue is not in draw()

• So if you click on the screen does it update your board as expected? The first thing is to add a simple println("Clicked"); in mousePressed(). Now you see that your click event is working well since the string is shown each time you click, so it must be an issue with how you get your x and y.

• Use println(x); and println(y); in mousePressed() to have an idea of your values. You will notice that you don't get integers for x you get decimal number.

And this is your issue: you can not access board[1.543] it doesn't make sense, so you need to keep only the integer part of mouseX / bs. To do that you can use int() like this:

First of all, you are not using mysqli properly. Please learn about prepared statements.

To get the data from the database you can use the following code:

Replaced this: