Source code for tigercontrol.experiments.metrics

# Metrics for evaluation

import jax.numpy as np 

def mse(y_pred, y_true):
    ''' Description: mean-square-error loss

        Args:
            y_pred : value predicted by model
            y_true : ground truth value
            eps: some scalar
    '''
    return np.sum((y_pred - y_true)**2)
    
def cross_entropy(y_pred, y_true, eps=1e-9):
    ''' Description: cross entropy loss, y_pred is equivalent to logits and y_true to labels
    
        Args:
            y_pred : value predicted by model
            y_true : ground truth value
            eps: some scalar
    '''
    return - np.dot(y_true, np.log(y_pred + eps))