Github user iyerr3 commented on the issue:
https://github.com/apache/madlib/pull/247
@kaknikhil
Yes, both those changes are related to MLP and ideally should have been in
a different commit. We can still make it that way by reverting them here and
then reintroducing these changes.
The `1.` is for not copying the coefficients every time we rebind and `2.`
is for moving the coefficient initialization from python to c++, since the
model is a c++ variable. They're both optimizations not necessarily required
immediately.
My suggestion is that we keep the changes here and avoid a redundant
commit. We've another task of fixing MLP's initialization (jira pending) - as
part of that ticket we can also utilize the `initialize` function. What do you
think?
---
