metrics.py

Go to the documentation of this file.

from tensorflow.python.ops import math_ops
import tensorflow as tf 
 
 
class MSA(tf.keras.metrics.Mean):
    ''' Mean Symmetric Accuracy '''
 
    def __init__(self, extract_estimate=lambda x:x, invert_scaling=lambda x:x):
        super(MSA, self).__init__(name='MSA', dtype=float)
        self._extract = extract_estimate
        self._invert  = invert_scaling
 
    def update_state(self, y_true, y_pred, *args, **kwargs):
        y_true = self._invert(y_true)
        y_pred = self._invert( self._extract(y_pred) )
        value  = tf.math.abs( tf.math.log(y_pred / y_true) )
        return super().update_state(value, *args, **kwargs) 
 
    def result(self):
        return 100. * (tf.math.exp(math_ops.div_no_nan(self.total, self.count)) - 1.)