core.utils.metrics

def accuracy(tp, fp, fn, tn):
	"""
		Returns accuracy as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return (tp+tn)/(tp + fp + fn + tn) if (tp + fp + fn + tn) > 0 else -1

def precision(tp, fp, fn, tn):
	"""
		Returns precision as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return tp/(tp+fp) if (tp+fp) > 0 else -1

def recall(tp, fp, fn, tn):
	"""
		Returns recall as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return tp/(tp+fn) if (tp+fn) > 0 else -1

def f1(tp, fp, fn, tn):
	"""
		Returns f1-score as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return (2*tp)/(2*tp + fp + fn) if (2*tp + fp + fn) > 0 else -1

def calc_metrics(labels, predictions):
	"""
		Calculate the metrics accuracy, precision, recall, and f1-score. 

		Args:
			labels (list): The list of the correct labels (int or bool)
			predictions (list): The list of the predicted labels (int or bool)

		Returns:
			dictionary containing calculates values
	"""
	tp, fp, fn, tn = 0, 0, 0, 0
	for l,p in zip(labels, predictions):
		if l and p:
			tp += 1
		elif l and not p:
			fn += 1
		elif not l and p:
			fp += 1
		else:
			tn += 1

	return {
		'accuracy': accuracy(tp, fp, fn, tn),
		'precision': precision(tp, fp, fn, tn),
		'recall': recall(tp, fp, fn, tn),
		'f1': f1(tp, fp, fn, tn),
		'num' : len(predictions), 
		'num_positive_labels' : labels.sum(),
		'num_negative_labels' : ((labels - 1) * -1).sum(),
		'num_positive_predictions' : predictions.sum(),
		'num_negative_predictions' : ((predictions - 1) * -1).sum()
	}


def _interval_similarity_or(start_i, end_i, start_j, end_j):	
	if start_j <= start_i and end_j >= end_i: # [ j -- ( i == i ) -- j ] 
		i_intersect = end_i - start_i + 1
	elif start_i <= start_j and end_i >= end_j: # ( i -- [ j == j ] -- i )
		i_intersect = end_j - start_j + 1
	elif start_i <= start_j and end_i <= end_j and end_i >= start_j: # ( i -- [ j == i ) -- j ]
		i_intersect = end_i - start_j + 1
	elif start_i >= start_j and end_i >= end_j and end_j >= start_i: # [ j -- ( i == j ] -- i )
		i_intersect = end_j - start_i + 1
	else: # no overlap
		i_intersect = 0
	
	return i_intersect / (end_i - start_i + 1)
	

def interval_similarity(start_i, end_i, start_j, end_j):
	"""
		Computes the similarity between two intervals (of natural numbers).

		Args:
			start_i (int): The start of interval i
			end_i (int): The end of interval i
			start_j (int): The start of interval j
			end_j (int): The end of interval j
		Returns:
			float in [0-1]  

			S_{OR}([start_i, end_i], [start_j, end_j])

		See [Novel similarity measure for interval-valued data based on overlapping ratio](https://ieeexplore.ieee.org/document/8015623)
		for the paper ([PDF](https://www.researchgate.net/publication/318311193_Novel_similarity_measure_for_interval-valued_data_based_on_overlapping_ratio)).
		Uses the [T_min Norm](https://en.wikipedia.org/wiki/T-norm#Prominent_examples)
	"""
	if start_i > end_i:
		start_i, end_i = end_i, start_i
	if start_j > end_j:
		start_j, end_j = end_j, start_j

	return min(
		_interval_similarity_or(start_i, end_i, start_j, end_j),
		_interval_similarity_or(start_j, end_j, start_i, end_i)
	)

def accuracy(tp, fp, fn, tn):
	"""
		Returns accuracy as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return (tp+tn)/(tp + fp + fn + tn) if (tp + fp + fn + tn) > 0 else -1

def precision(tp, fp, fn, tn):
	"""
		Returns precision as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return tp/(tp+fp) if (tp+fp) > 0 else -1

def recall(tp, fp, fn, tn):
	"""
		Returns recall as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return tp/(tp+fn) if (tp+fn) > 0 else -1

def f1(tp, fp, fn, tn):
	"""
		Returns f1-score as float given true-positives ``tp``, false-positives ``fp``,
		false-negatives ``fn``, true-negatives ``tn``.
	"""
	return (2*tp)/(2*tp + fp + fn) if (2*tp + fp + fn) > 0 else -1

def calc_metrics(labels, predictions):
	"""
		Calculate the metrics accuracy, precision, recall, and f1-score. 

		Args:
			labels (list): The list of the correct labels (int or bool)
			predictions (list): The list of the predicted labels (int or bool)

		Returns:
			dictionary containing calculates values
	"""
	tp, fp, fn, tn = 0, 0, 0, 0
	for l,p in zip(labels, predictions):
		if l and p:
			tp += 1
		elif l and not p:
			fn += 1
		elif not l and p:
			fp += 1
		else:
			tn += 1

	return {
		'accuracy': accuracy(tp, fp, fn, tn),
		'precision': precision(tp, fp, fn, tn),
		'recall': recall(tp, fp, fn, tn),
		'f1': f1(tp, fp, fn, tn),
		'num' : len(predictions), 
		'num_positive_labels' : labels.sum(),
		'num_negative_labels' : ((labels - 1) * -1).sum(),
		'num_positive_predictions' : predictions.sum(),
		'num_negative_predictions' : ((predictions - 1) * -1).sum()
	}

def interval_similarity(start_i, end_i, start_j, end_j):
	"""
		Computes the similarity between two intervals (of natural numbers).

		Args:
			start_i (int): The start of interval i
			end_i (int): The end of interval i
			start_j (int): The start of interval j
			end_j (int): The end of interval j
		Returns:
			float in [0-1]  

			S_{OR}([start_i, end_i], [start_j, end_j])

		See [Novel similarity measure for interval-valued data based on overlapping ratio](https://ieeexplore.ieee.org/document/8015623)
		for the paper ([PDF](https://www.researchgate.net/publication/318311193_Novel_similarity_measure_for_interval-valued_data_based_on_overlapping_ratio)).
		Uses the [T_min Norm](https://en.wikipedia.org/wiki/T-norm#Prominent_examples)
	"""
	if start_i > end_i:
		start_i, end_i = end_i, start_i
	if start_j > end_j:
		start_j, end_j = end_j, start_j

	return min(
		_interval_similarity_or(start_i, end_i, start_j, end_j),
		_interval_similarity_or(start_j, end_j, start_i, end_i)
	)