Module monk.gluon.losses.retrieve_loss

Functions

def retrieve_loss(system_dict)

Retrieve loss post state changes

Args

system_dict : dict

System dictionary storing experiment state and set variables

Returns

dict

updated system dict

Expand source code

def retrieve_loss(system_dict):
        '''
    Retrieve loss post state changes

    Args:
        system_dict (dict): System dictionary storing experiment state and set variables

    Returns:
        dict: updated system dict
    '''
        system_dict["local"]["criterion"] = system_dict["hyper-parameters"]["loss"]["name"];
        name = system_dict["local"]["criterion"];

        if(name == "l1"):
                system_dict["local"]["criterion"] = mx.gluon.loss.L1Loss(
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);

        elif(name == "l2"):
                system_dict["local"]["criterion"] = mx.gluon.loss.L2Loss(
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);

        elif(name == "softmaxcrossentropy"):
                system_dict["local"]["criterion"] = mx.gluon.loss.SoftmaxCrossEntropyLoss(
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"],
                        axis=system_dict["hyper-parameters"]["loss"]["params"]["axis_to_sum_over"],
                        sparse_label=system_dict["hyper-parameters"]["loss"]["params"]["label_as_categories"],
                        from_logits=False);

        elif(name == "crossentropy"):
                system_dict["local"]["criterion"] = mx.gluon.loss.SoftmaxCrossEntropyLoss(
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"],
                        axis=system_dict["hyper-parameters"]["loss"]["params"]["axis_to_sum_over"],
                        sparse_label=system_dict["hyper-parameters"]["loss"]["params"]["label_as_categories"],
                        from_logits=True);

        elif(name == "sigmoidbinarycrossentropy"):
                system_dict["local"]["criterion"] = mx.gluon.loss.SigmoidBinaryCrossEntropyLoss(
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"],
                        from_sigmoid=False);


        elif(name == "binarycrossentropy"):
                system_dict["local"]["criterion"] = mx.gluon.loss.SigmoidBinaryCrossEntropyLoss(
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        from_sigmoid=True);


        elif(name == "kldiv"):
                system_dict["local"]["criterion"] = mx.gluon.loss.KLDivLoss(
                        from_logits=system_dict["hyper-parameters"]["loss"]["params"]["log_pre_applied"], 
                        axis=system_dict["hyper-parameters"]["loss"]["params"]["axis_to_sum_over"], 
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"], 
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);

        elif(name == "poissonnll"):
                system_dict["local"]["criterion"] = mx.gluon.loss.PoissonNLLLoss(
                        from_logits=system_dict["hyper-parameters"]["loss"]["params"]["log_pre_applied"],
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);

        elif(name == "huber"):
                system_dict["local"]["criterion"] = mx.gluon.loss.HuberLoss(
                        rho=system_dict["hyper-parameters"]["loss"]["params"]["threshold_for_mean_estimator"],
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);

        elif(name == "hinge"):
                system_dict["local"]["criterion"] = mx.gluon.loss.HingeLoss(
                        margin=system_dict["hyper-parameters"]["loss"]["params"]["margin"],
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);

        elif(name == "squaredhinge"):
                system_dict["local"]["criterion"] = mx.gluon.loss.SquaredHingeLoss(
                        margin=system_dict["hyper-parameters"]["loss"]["params"]["margin"],
                        weight=system_dict["hyper-parameters"]["loss"]["params"]["weight"],
                        batch_axis=system_dict["hyper-parameters"]["loss"]["params"]["batch_axis"]);    

        return system_dict;