How much "hand-tweaking" is allowed?
22 views
Skip to first unread message
Danny
Feb 25, 2012, 5:31:00 PM2/25/12
to Machine March Madness
Scott writes...
You might also want to clarify how much "handwork" is allowed. For
example, I have an algorithm that tries to force a certain number of
"upsets" in the first round regardless of what my base model predicts.
My response:
It's fine to run some other procedure on top of some base model's
predictions, but the other procedure should be some fixed algorithm
(e.g., pick at least N upsets; or, if the model predicts a near-tie,
choose the upset for up to the 10 closest games; or fancier variations
on these).
You might also want to clarify how much "handwork" is allowed. For
example, I have an algorithm that tries to force a certain number of
"upsets" in the first round regardless of what my base model predicts.
My response:
It's fine to run some other procedure on top of some base model's
predictions, but the other procedure should be some fixed algorithm
(e.g., pick at least N upsets; or, if the model predicts a near-tie,
choose the upset for up to the 10 closest games; or fancier variations
on these).
Dr. Pain
Feb 25, 2012, 7:21:59 PM2/25/12
to Machine March Madness
On Feb 25, 5:31 pm, Danny <dannytar...@gmail.com> wrote:
> It's fine to run some other procedure on top of some base model's
> predictions, but the other procedure should be some fixed algorithm
> (e.g., pick at least N upsets; or, if the model predicts a near-tie,
> choose the upset for up to the 10 closest games; or fancier variations
> on these).
That's my take; I'm fine with hand-implemented algorithm as long as it
> It's fine to run some other procedure on top of some base model's
> predictions, but the other procedure should be some fixed algorithm
> (e.g., pick at least N upsets; or, if the model predicts a near-tie,
> choose the upset for up to the 10 closest games; or fancier variations
> on these).
does not involve human judgement.
-- Scott
Reply all
Reply to author
Forward
0 new messages