Supercollider: write and read MLP models

Usage Questions
1

Hello all,

Might just be that I haven’t looked hard enough but I’m having trouble saving and restoring models created with the FluidMLPRegressor or FluidMLPClassifier in SuperCollider. The .write method seems to print the model to a file. I assumed I could restore the model with .read but got an error which refers to an ID, e.g. ‘ERROR: FluidMLPRegressor no instance with ID 4’. Am I on the right track here; is there any documentation on this that I have overlooked?

Cheers
Erik

2

Hi @Erik and welcome,

That sounds like an error I thought we got rid of in the last release. Does the following fail for you?

(
//just put anything into a dataset so we can 'fit' the MLP before writing
~ds = FluidDataSet(s);
~buf = Buffer.alloc(s,1,1);
~ds.setPoint(\1,~buf); 
//mlp1 we'll fit then save
~mlp1 = FluidMLPRegressor(s);
~mlp1.fit(~ds,~ds);
~mlp1.write("/tmp/mlptest.json"); 
//read into a 2nd mlp
~mlp2 = FluidMLPRegressor(s); 
~mlp2.read("/tmp/mlptest.json"); 
)

and does this work?

(
fork{
    //just put anything into a dataset so we can 'fit' the MLP before writing
    ~ds = FluidDataSet(s);
    s.sync; 
    ~buf = Buffer.alloc(s,1,1);
    ~ds.setPoint(\1,~buf); 
    //mlp1 we'll fit then save
    ~mlp1 = FluidMLPRegressor(s);
    s.sync; 
    ~mlp1.fit(~ds,~ds);
    ~mlp1.write("/tmp/mlptest.json"); 
    //read into a 2nd mlp
    ~mlp2 = FluidMLPRegressor(s); 
    s.sync; 
    ~mlp2.read("/tmp/mlptest.json"); 
}
)

If so, maybe an update is indicated. If you do

FluidDataSet.version

it should print your version to the console: current is version 1.0.0-TB2.beta5

3

Hi Owen!
Thanks a lot for this. Yes I did get errors in the first block of code but not the second. Also turns out I have version 1.0.0-TB2.beta4 so I need to update. However, I did get it to work by creating the instance first and then calling .read, rather than doing it all in one line.

Erik

4

All works fine after upgrade.