Depending on how (ip2,ip3) pairs vary, the intermediate variable may have one too many dimensions (both ip2 and ip3_values), where there is only actually data available along the diagonal.
E.g. for some datasets a 1-hour accumulation would be encoded with (ip2,ip3) pairs of (1,2), (2, 3), ... (48,47) giving an outer product of ip2 axis (1..48) and ip3 axis (0..47), but no records defined except along the main diagonal.
This gets cleaned up further in the accum logic which will squash it into a single dimension, but not before a misleading warning message is printed.
Depending on how (ip2,ip3) pairs vary, the intermediate variable may have one too many dimensions (both ip2 and ip3_values), where there is only actually data available along the diagonal.
E.g. for some datasets a 1-hour accumulation would be encoded with (ip2,ip3) pairs of (1,2), (2, 3), ... (48,47) giving an outer product of ip2 axis (1..48) and ip3 axis (0..47), but no records defined except along the main diagonal.
This gets cleaned up further in the accum logic which will squash it into a single dimension, but not before a misleading warning message is printed.