Alphago implementing

Sorry if this is a stupid question, but I was wondering why nobody has tried to implement alphago’s moves. there is kifu on alphago 60 online games five games against Lee sedol. 3 games against ke jie if I am not mistaken and quite a lot against itself so i am wondering why nobody has tried to implement alphago’s moves in a neural network. Again I apologize if this is not possible and is a stupid question, I am new to neural networks and related stuff.

The original code for AlphaGo and its variants (i.e., Zero) have been kept as closed-source by Google/DeepMind.

However, there have been various open-source efforts to essentially re-implement the algorithms used by these bots. Here are two prominent examples:

You’re misunderstanding what I am trying to say, I mean implementing the moves of alphago’s kifu,

Perhaps, I am.
I don’t know what you mean by “implementing the moves of alphago’s kifu”

The original AlphaGo algorithm and its variants are implemented using neural networks.

As I am new to stuff like this, ‘‘implementing’’ may be the wrong word, I mean putting alphago’s move into a go ai is what I mean in simple terms.

I think you might have a fundamental misunderstanding about what AlphaGo is. You seem to be asking about things that have already been done.

AlphaGo is a go AI, which was built using neural networks.

Many others have already built other go AIs, again using neural networks, by following the pattern/design of AlphaGo, which was described in published scientific papers.

No, I simply do not mean that, I mean Implementing alphago moves into the ai because i seen them implement professional moves so why couldn’t they implement alphago’s moves into the ai, I don’t mean copying alphago pattern or anything i mean implementing alphago’s moves, I already read nature and nothing has been suggested that it’s not possible to implement alphago moves into a ai, I don’t really understand what u mean but nor do u seem to understand what i mean it’s confusing.

What exactly do you mean by an AI? If you mean a computer program that leverages neural networks to play go, then that is exactly what AlphaGo already is.

I think you might ultimately be asking if whether another go program could be trained to more closely emulate AlphaGo by using just its very limited set of released game records. The answer to that question is probably no, since a few hundred games is simply an extremely small dataset to work with when training neural networks of the scale and complexity that are involved.

2 Likes

I mean by implementing alphagos move along with other professional kifu? and yes your latter sentence is what i mean.

If you rephrase the question: “Has anybody trained a neural network from AlphaGo games and professional games,” then the answer is yes. Leela Master is a neural network “weight” trained from AlphaGo games and professional games. The dataset used is found here.

1 Like

‘‘Leela Master weight is training from leela zero self-play sgf and human sgf file’’, Leela master doesn’t say anything about training from alphago games??

You didn’t read the link.

“The human sgf is mainly download from https://github.com/yenw/computer-go-dataset

That dataset contains AlphaGo games.

Ok, but how do you even put the weight into lizzie folder? sorry I am truly inexperienced with ai stuff.

It’s not a problem. I assume you already installed Lizzie and have it working. If that is the case, then find the directory where it’s installed, and find the file “network.gz” in that folder. You’re going to replace that file with another network you download. You can handle this in one of two ways: (1) call it “network.gz” and move the old file out, or (2) call it a different name and then tell Lizzie’s config file to use the new network instead.

Um, I get what you mean there’s just one problem do you mean put the network weight into the network folder?

Yes, you’ll put the new network in the Lizzie root folder—wherever the “network.gz” file is. It’s the same folder as “lizzie.jar” and “config.txt.”

When I do this it says “Winrar can only add rar or win files”.

Yo buddy, you posted with the wrong account.

4 Likes

Then don’t open the network.gz file with Winrar. The network file is for Lizzie not for you. All you need to do is put the network file in Lizzie’s directory, make sure config.txt is calling it up, and then open Lizzie.

you mean like the 3-3 invasion? lots of people are doing that and lots of pros have learnt a lot from the alphago games.

if not what exactly do you mean by not implementing its move?