wehub-resource-sync
91 lines
2.2 KiB
Python
91 lines
2.2 KiB
Python
# coding:utf-8
|
|
import autograd.numpy as np
|
|
|
|
EPS = 1e-15
|
|
|
|
|
|
def unhot(function):
|
|
"""Convert one-hot representation into one column."""
|
|
|
|
def wrapper(actual, predicted):
|
|
if len(actual.shape) > 1 and actual.shape[1] > 1:
|
|
actual = actual.argmax(axis=1)
|
|
if len(predicted.shape) > 1 and predicted.shape[1] > 1:
|
|
predicted = predicted.argmax(axis=1)
|
|
return function(actual, predicted)
|
|
|
|
return wrapper
|
|
|
|
|
|
def absolute_error(actual, predicted):
|
|
return np.abs(actual - predicted)
|
|
|
|
|
|
@unhot
|
|
def classification_error(actual, predicted):
|
|
return (actual != predicted).sum() / float(actual.shape[0])
|
|
|
|
|
|
@unhot
|
|
def accuracy(actual, predicted):
|
|
return 1.0 - classification_error(actual, predicted)
|
|
|
|
|
|
def mean_absolute_error(actual, predicted):
|
|
return np.mean(absolute_error(actual, predicted))
|
|
|
|
|
|
def squared_error(actual, predicted):
|
|
return (actual - predicted) ** 2
|
|
|
|
|
|
def squared_log_error(actual, predicted):
|
|
return (np.log(np.array(actual) + 1) - np.log(np.array(predicted) + 1)) ** 2
|
|
|
|
|
|
def mean_squared_log_error(actual, predicted):
|
|
return np.mean(squared_log_error(actual, predicted))
|
|
|
|
|
|
def mean_squared_error(actual, predicted):
|
|
return np.mean(squared_error(actual, predicted))
|
|
|
|
|
|
def root_mean_squared_error(actual, predicted):
|
|
return np.sqrt(mean_squared_error(actual, predicted))
|
|
|
|
|
|
def root_mean_squared_log_error(actual, predicted):
|
|
return np.sqrt(mean_squared_log_error(actual, predicted))
|
|
|
|
|
|
def logloss(actual, predicted):
|
|
predicted = np.clip(predicted, EPS, 1 - EPS)
|
|
loss = -np.sum(actual * np.log(predicted))
|
|
return loss / float(actual.shape[0])
|
|
|
|
|
|
def hinge(actual, predicted):
|
|
return np.mean(np.max(1.0 - actual * predicted, 0.0))
|
|
|
|
|
|
def binary_crossentropy(actual, predicted):
|
|
predicted = np.clip(predicted, EPS, 1 - EPS)
|
|
return np.mean(
|
|
-np.sum(actual * np.log(predicted) + (1 - actual) * np.log(1 - predicted))
|
|
)
|
|
|
|
|
|
# aliases
|
|
mse = mean_squared_error
|
|
rmse = root_mean_squared_error
|
|
mae = mean_absolute_error
|
|
|
|
|
|
def get_metric(name):
|
|
"""Return metric function by name"""
|
|
try:
|
|
return globals()[name]
|
|
except Exception:
|
|
raise ValueError("Invalid metric function.")
|