niklib.models.evaluators.metrics package

Module contents

Different metrics from what sklearn.metrics offers or our own custom metrics

Note that for deep learning based metrics that need computational graph or neural network API such as using another model or a network as a metric (similar to feature extraction), then user should first implement all required models/networks in niklib.models.estimators.networks or other appropriate submodules then import them here and just make a call to them. For instance assume we have VGG models in niklib.models.estimators.networks.vgg.py:

from niklib.models.estimators.networks.vgg import vgg16
def vgg16_metric_via_layer_x_and_y(y_true, y_pred):
    ...
    value_1 = vgg16(y_true, y_pred, layer=1)
    value_2 = vgg16(y_true, y_pred, layer=2)
    value = value_1 + value_2
    ...
    return value