Closed Azhag closed 8 years ago
Modify how the Fisher information from curvature is computed.
Shouldn't be a posterior weighted integral over the entire stimulus space, but only the curvature at the ML value.
EDIT: this is #23
FI theo -> FI theo large N seems weird, test again.
There was an error in FI, wasn't estimated at exact stimulus value.
Now doing it properly.
Mismatch now for Covariance KL, but putting it into another issue.
Modify how the Fisher information from curvature is computed.
Shouldn't be a posterior weighted integral over the entire stimulus space, but only the curvature at the ML value.
EDIT: this is #23