[javacpp-pytorch] OptimizerParamGroupVector class need have push_back method to insert data

[mullerhai]

Hi @saudet

  now. OptimizerParamGroupVector can not insert data , but in fact. we need use it  put Optimizer param data 

        TensorVector decayParams   = new TensorVector();
        TensorVector noDecayParams = new TensorVector();

        StringTensorDict named = model.named_parameters();
        StringTensorDictItemVector.Iterator it = named.begin();
        while (!it.equals(named.end())) {
            Tensor p = it.get().access();
            if (p.requires_grad()) {
                if (p.dim() >= 2) decayParams.push_back(p);
                else              noDecayParams.push_back(p);
            }
            it.increment();
        }

        // Build two parameter groups
        AdamWOptions baseOpts = new AdamWOptions(lr);
        baseOpts.betas().put(0, beta1);
        baseOpts.betas().put(1, beta2);
        baseOpts.weight_decay().put(weightDecay);

        AdamWOptions noDecayOpts = new AdamWOptions(lr);
        noDecayOpts.betas().put(0, beta1);
        noDecayOpts.betas().put(1, beta2);
        noDecayOpts.weight_decay().put(0.0);

       OptimizerParamGroupVector groups = new OptimizerParamGroupVector();
       groups.push_back(new OptimizerParamGroup(decayParams,   baseOpts));
       groups.push_back(new OptimizerParamGroup(noDecayParams, noDecayOpts));
       return new AdamW(groups, baseOpts);


//        TensorVector groups = new TensorVector();
//        groups.push_back(new OptimizerParamGroup(decayParams,   baseOpts));
//        groups.push_back(new OptimizerParamGroup(noDecayParams, noDecayOpts));

the original code

public class AdamW extends Optimizer {
    public AdamW(Pointer p) {
        super(p);
    }

    public AdamW(@Const @ByRef OptimizerParamGroupVector param_groups, @ByVal(nullValue = "torch::optim::AdamWOptions{}") AdamWOptions defaults) {
        super((Pointer)null);
        this.allocate(param_groups, defaults);
    }

    private native void allocate(@Const @ByRef OptimizerParamGroupVector var1, @ByVal(nullValue = "torch::optim::AdamWOptions{}") AdamWOptions var2);

    public AdamW(@Const @ByRef OptimizerParamGroupVector param_groups) {
        super((Pointer)null);
        this.allocate(param_groups);
    }

[saudet]

Done in commit 33cd4d8. Please try again with the snapshots:
https://central.sonatype.com/service/rest/repository/browse/maven-snapshots
https://central.sonatype.org/publish/publish-portal-snapshots/#consuming-snapshot-releases-for-your-project

[mullerhai]

Done in commit 33cd4d8. Please try again with the snapshots: https://central.sonatype.com/service/rest/repository/browse/maven-snapshots https://central.sonatype.org/publish/publish-portal-snapshots/#consuming-snapshot-releases-for-your-project

I have double check ，these **Vector Class also not have put or push_back method,
FunctionSchemaVector
DataPtrVector
FunctionPreHookVector
FunctionPostHookVector
PropertyVector
StringAnyModuleDictItemVector
StringSharedModuleDictItemVector
StringTensorDictItemVector
WeakStorageVector

[saudet]

Do you need any of them?

[mullerhai]

Do you need any of them?

I'm not entirely sure. These nine vectors are the only remaining ones that haven't had the put and push_back operations implemented yet. They are used in advanced differential autograd equations, some in advanced distributed training, some point to storage, and others are involved in traversing lists and dictionaries. If I had to suggest, I think it would be convenient to first try implementing the insert method for the following classes:
FunctionSchemaVector
DataPtrVector
FunctionPreHookVector
FunctionPostHookVector
PropertyVector
StringSharedModuleDictItemVector
If possible, we can implement the insert method for all nine of them. Thanks for your hard work.