need more metrics(mse, ridge) for regression

[songhappy]

Currently, it only has "mae" for regression, "mse" and "ridge" are needed at least.

def to_bigdl_metric(metric):
    metric = metric.lower()
    if metric == "accuracy" or metric == "acc":
        return metrics.Accuracy()
    elif metric == "top5accuracy" or metric == "top5acc":
        return metrics.Top5Accuracy()
    elif metric == "mae":
        from bigdl.optim.optimizer import MAE
        return MAE()
    elif metric == "auc":
        return metrics.AUC()
    elif metric == "loss":
        return Loss()
    elif metric == "treennaccuracy":
        return TreeNNAccuracy()
    else:
        raise TypeError("Unsupported metric: %s" % metric)

[hkvision]

Seems that in BigDL only MAE is wrapped as a metrics.

[hkvision]

Also, we may need to add some other metrics like the one used by crf for nlp-architect integration.

[yangw1234]

Maybe enable user to create custom metrics (similar to cusomloss except some metrics may require a custom reducer)?