[BUG] Min Trace method="mint_shrink"

[AzulGarza]

Users have reported issues trying to run MinTrace(method=='mint_shrink').

I suspect that it is related to datasets with time series of distinct lengths (in our examples all time series have the same length and the method works fine). In particular, I think that line 441 below is not working correctly. It should ignore the NAS in the residuals matrix generated by the difference in lengths, but perhaps it is not ignoring them.

hierarchicalforecast/hierarchicalforecast/methods.py

Lines 438 to 457 in 2df1dd8

	residuals = (y_insample - y_hat_insample).T
	n, _ = residuals.shape
	masked_res = np.ma.array(residuals, mask=np.isnan(residuals))
	covm = np.ma.cov(masked_res, rowvar=False, allow_masked=True).data
	if method == 'wls_var':
	W = np.diag(np.diag(covm))
	elif method == 'mint_cov':
	W = covm
	elif method == 'mint_shrink':
	tar = np.diag(np.diag(covm))
	corm = cov2corr(covm)
	xs = np.divide(residuals, np.sqrt(np.diag(covm)))
	xs = xs[~np.isnan(xs).any(axis=1), :]
	v = (1 / (n * (n - 1))) * (crossprod(xs ** 2) - (1 / n) * (crossprod(xs) ** 2))
	np.fill_diagonal(v, 0)
	corapn = cov2corr(tar)
	d = (corm - corapn) ** 2
	lmd = v.sum() / d.sum()
	lmd = max(min(lmd, 1), 0)
	W = lmd * tar + (1 - lmd) * covm

From Linear-GitHub Sync

[AzulGarza]

According to the docs, it is the most likely reason since allow_masked=True simply propagates the mask but does not remove the nas.