Covariance calculation (between <LWA> and <U>) needs optimization about hn2016_falwa

[csyhuang]

Doing it row-wise seems to be feasible:

hn2016_falwa/falwa/stat_utils.py

Lines 5 to 21 in c909f0b

	def calculate_covariance(var_a, var_b):
	"""
	Calculate covariance of two variables in time.
	Args:
	var_a: a numpy array or handle that can access elements via [time, lat, lon]
	var_b: a numpy array or handle that can access elements via [time, lat, lon]
	Returns:
	cov_map in dimension of (lat, lon)
	"""
	lat_dim = var_a.shape[1]
	lon_dim = var_a.shape[2]
	cov_map = np.zeros((lat_dim, lon_dim))
	for j in range(lat_dim): # has to loop through a dimension to conserve memory
	cov_matrix = np.cov(m=var_a[:, j, :], y=var_b[:, j, :], rowvar=False)
	row_cov = np.diagonal(cov_matrix, offset=lon_dim)
	cov_map[j, :] = row_cov
	return cov_map

This issue is considered closed by 38ed52b