|
Nagys' tipple around half of a bad lot will give you one of our own (7)
We can’t stop here, this is bat country - Hunter S Thompson RIP
|
|
|
|
|
I get the idea - you don't want to do it tomorrow! 
Griffin, by any chance?
Bad command or file name. Bad, bad command! Sit! Stay! Staaaay...
|
|
|
|
|
Serve ya right for naggin
We can’t stop here, this is bat country - Hunter S Thompson RIP
modified 15-Aug-16 8:14am.
|
|
|
|
|
GIN around half of "RIFFRAFF" - oh dear...
But I guess I'm up tomorrow.
Bad command or file name. Bad, bad command! Sit! Stay! Staaaay...
|
|
|
|
|
Yep - naggin wasn't too bad either 
We can’t stop here, this is bat country - Hunter S Thompson RIP
|
|
|
|
|
You forgot V. and MQOTD! 
Anything that is unrelated to elephants is irrelephant
Anonymous
-
The problem with quotes on the internet is that you can never tell if they're genuine
Winston Churchill, 1944
-
I'd just like a chance to prove that money can't make me happy.
Me, all the time
|
|
|
|
|
Aint V. on vacation?
Rules for the FOSW ![ ^]
if(this.signature != "")
{
MessageBox.Show("This is my signature: " + Environment.NewLine + signature);
}
else
{
MessageBox.Show("404-Signature not found");
}
|
|
|
|
|
Really? I thought he'd just come back from vacation... 
Anything that is unrelated to elephants is irrelephant
Anonymous
-
The problem with quotes on the internet is that you can never tell if they're genuine
Winston Churchill, 1944
-
I'd just like a chance to prove that money can't make me happy.
Me, all the time
|
|
|
|
|
Well if he's not on vacation, just remember he sometimes doesn't post it 
Rules for the FOSW ![ ^]
if(this.signature != "")
{
MessageBox.Show("This is my signature: " + Environment.NewLine + signature);
}
else
{
MessageBox.Show("404-Signature not found");
}
|
|
|
|
|
The other day I was playing with the idea of a generic processor emulator. Would someone like to join the experiment to implement it faster? Here's what I am up to.
1) I am going to produce a gigabyte of random bytes ( I know, I know, you volunteer for that )
2) I am going to treat these bytes as Z80 op-codes,
3) I am going to feed them into neural network,
4) I am going to interpret them with an emulator and use expected processor behavior (i.e. state of registers, pc, flags, memory location read/writes, stack read/writes) to train the network,
5) I am going to wait for a week.
And if I am lucky I am going to end up with Z80 neural network. But in reality /if it works/ I am going to end up with a generic interpreter generator / duplicator.
...and with it we are going to start "the era of neuro-device" where a software is just a bunch of weights (and infos about the network structure)...a trained puppy, really...muwhahaha...
modified 15-Aug-16 4:45am.
|
|
|
|
|
Um...wouldn't it be better to start from a known-working Z80 program, rather than random data? If you're trying to train it to work as a Z80 then starting from stuff that only includes valid Z80 opcodes has got to be better than stuff that will immediately crash if fed to a real Z80?
Bad command or file name. Bad, bad command! Sit! Stay! Staaaay...
|
|
|
|
|
Actually all 8 bit codes are valid Z80 codes - not that those will necessary make a valid program...
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
Not all of them - the ED extension codes don't define an operation for some values.
Bad command or file name. Bad, bad command! Sit! Stay! Staaaay...
|
|
|
|
|
Decoding Z80 Opcodes[^]
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
That could be actually the most interesting part of a self-learning Z80 emulator...Will it learn to handle illegal opcode sequences as error and crash, or will behave like a true Z80 - wait and continue...
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
Funnily enough, random data would probably serve better, especially if backprop were used in place of the neural network.
With AI, the best starting point is often not the one that makes logical (programming-type) sense, because it has to learn how to work processes out for itself, not how to take data and reprocess it.
I wanna be a eunuchs developer! Pass me a bread knife!
|
|
|
|
|
Tomaž Štih wrote: 2) I am going to treat these bytes as Z80 op-codes,
Who will threat them like that? The ANN? Some pre-interpreter? Not clear at what point those codes will have meaning...
Tomaž Štih wrote: 3) I am going to feed them into neural network,
What? The byte or it's meaning?
Tomaž Štih wrote: 4) I am going to interpret them with an emulator and use expected processor behavior (i.e. state of registers, pc, flags, memory location read/writes, stack read/writes) to train the network,
What will you interpret? The output of the feed? And what exactly will be the training data after all this? The op-codes? The interpreted data (whatever that data is)?
And most importantly - you should state what kind of collaborations you are looking for!
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
Kornfeld Eliyahu Peter wrote: Who will threat them like that? The ANN? Some pre-interpreter? Not clear at what point those codes will have meaning...
Input bytes should be raw data. No interpreting. One could try feeding the classic net with 24 inputs (3 bytes) at a time and doing the convolution (sliding 1 byte instead of 3 bytes) across the raw data to enable long instructions. Or one could try a recurring neural network.Kornfeld Eliyahu Peter wrote: What? The byte or it's meaning?
The byte.Kornfeld Eliyahu Peter wrote: What will you interpret? The output of the feed? And what exactly will be the training data after all this? The op-codes? The interpreted data (whatever that data is)?
Neural network structure -
Input layer: op-code stream
Hidden layer: no idea. 
Output layer: task to execute
- one output for address write
- one output for address read
- one output for register write
- one output for register read
- 16 outputs for data
- 16 outputs for address
- 3 outputs for register 1
- 3 outputs for register 2
- one output for each flag
...
So if the instruction is LD A,10. Then it triggers register write with lower outputs filled with 10. If it is POP DE then it triggers stack read and the registers are D and E. So basically interpreting commands and reducing them to simpler commands ( hmm, well, actually almost the same commands; of course one may end up with decoding logic that is way more complex then the actual logic, but it is just weights so it is in a way "learned" ).
Training is done with help of an actual z80 emulator (an existing c# code can be used). The emulator should interpret the input bytes and set correct outputs. These should then be compared to the ones produced by our NN and used for back-propagation.Kornfeld Eliyahu Peter wrote: And most importantly - you should state what kind of collaborations you are looking for!
Any sort of programming/theoretical cooperation is fine. I'm new to neural networks. I mastered classic neural nets and am now into convolution neural nets, but could use some help in understanding recurring neural networks. I'm sure someone else could use something I've got. And a result can be a shared article for CP?
I'm interested in testing if the idea works and learning new stuff while doing it.
Sincerely,
T.
modified 15-Aug-16 6:29am.
|
|
|
|
|
Now I'm lost you even more...
Do you want to train an ANN from nothing to a Z80 interpreter or from a Z80 interpreter to a all-purpose interpreter?
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
1) Ok. For start I want to train ANN from nothing to Z80 interpreter. That's a limited, not to difficult task. It will provide a proof of concept and I will learn a lot about how to approach such problem.
2) Then, if concept is good and it works, it could be used to train to interpret, compile, optimize other programming languages.
|
|
|
|
|
That's more clear now...
All you have to understand, that 'nothing' means that there is no hidden layer in your ANN, and that means nothing will build as the ANN lack any connection that would change the input...
So first thing is to set up minimal rules, that can be the hearth of a Z80 interpreter (like one of the basic rules of the Z80 is that there is no such thing as unknown op-code)...
As you want to feed back the output errors as new training material, you have to have a second rule set (a parallel hidden layer) to start with from that point of view...
Skipper: We'll fix it.
Alex: Fix it? How you gonna fix this?
Skipper: Grit, spit and a whole lotta duct tape.
|
|
|
|
|
A bit like an infinite number of monkeys with typewriters. Let us know when it's finished.
|
|
|
|
|
That's the usual definition we use for AI procedures.
The thing is, we settle for one line of Shakespeare, and use billions of iterations, rather than infinite monkeys.
I wanna be a eunuchs developer! Pass me a bread knife!
|
|
|
|
|
Tomaž Štih wrote: I am going to produce a gigabyte of random bytes
Well, as the saying goes, "garbage in, garbage out" 
Marc
|
|
|
|
|
Isn't that the premise behind Man V Food?
This space for rent
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
