AdaBoost Question

[Mark]

I'm using AdaBoost on optical flow data to learn to recognize human
actions. I've split my optical flow data into 5 channels (positive x,
negative x, positive y, negative y, and 'zero'). I know how to use
AdaBoost on a single channel, but I'm not sure how to combine the 5
channels to form a better classifier. Do I run AdaBoost on each
channel, and then pass the 5 outputs to another AdaBoost, or should I
simply concatenate the 5 channels and run AdaBoost once on all of
them?

Thanks,
Mark

[Victor Bazarov]

Mark wrote:

I'm using AdaBoost on optical flow data to learn to recognize human
actions. I've split my optical flow data into 5 channels (positive x,
negative x, positive y, negative y, and 'zero'). I know how to use
AdaBoost on a single channel, but I'm not sure how to combine the 5
channels to form a better classifier. Do I run AdaBoost on each
channel, and then pass the 5 outputs to another AdaBoost, or should I
simply concatenate the 5 channels and run AdaBoost once on all of
them?

What's "AdaBoost"?

V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask

[Mark]

On May 14, 1:07 pm, Victor Bazarov <[e-mail]:

Mark wrote:

I'm using AdaBoost on optical flow data to learn to recognize human
actions. I've split my optical flow data into 5 channels (positive x,
negative x, positive y, negative y, and 'zero'). I know how to use
AdaBoost on a single channel, but I'm not sure how to combine the 5
channels to form a better classifier. Do I run AdaBoost on each
channel, and then pass the 5 outputs to another AdaBoost, or should I
simply concatenate the 5 channels and run AdaBoost once on all of
them?

What's "AdaBoost"?

V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask

AdaBoost is a boosting/machine learning algorithm. It's short for
Adaptive Boosting (http://en.wikipedia.org/wiki/AdaBoost). My
question isn't exactly specific to AdaBoost. The output of AdaBoost
is a class. For example, I've got some videos where I am waving my
hand, and some where I am not. AdaBoost spits out "pos" for waving,
or "neg" for not waving. I could run AdaBoost on my 5 channels, and
do something like a majority vote, or use AdaBoost to figure out how
many "pos"s it actually needs.... or I can perhaps pass the confusion
matrix in as input... Basically, I need to figure out how to make it
multi-layer. (Btw, I'm achieving a 13% error rate using just 1
channel! This is exciting)

Mark

[Mark]

On May 14, 1:14 pm, Mark <[e-mail]:

On May 14, 1:07 pm, Victor Bazarov <[e-mail]:

Mark wrote:

I'm using AdaBoost on optical flow data to learn to recognize human
actions. I've split my optical flow data into 5 channels (positive x,
negative x, positive y, negative y, and 'zero'). I know how to use
AdaBoost on a single channel, but I'm not sure how to combine the 5
channels to form a better classifier. Do I run AdaBoost on each
channel, and then pass the 5 outputs to another AdaBoost, or should I
simply concatenate the 5 channels and run AdaBoost once on all of
them?

What's "AdaBoost"?

V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask

AdaBoost is a boosting/machine learning algorithm. It's short for
Adaptive Boosting (http://en.wikipedia.org/wiki/AdaBoost). My
question isn't exactly specific to AdaBoost. The output of AdaBoost
is a class. For example, I've got some videos where I am waving my
hand, and some where I am not. AdaBoost spits out "pos" for waving,
or "neg" for not waving. I could run AdaBoost on my 5 channels, and
do something like a majority vote, or use AdaBoost to figure out how
many "pos"s it actually needs.... or I can perhaps pass the confusion
matrix in as input... Basically, I need to figure out how to make it
multi-layer. (Btw, I'm achieving a 13% error rate using just 1
channel! This is exciting)

Mark

Actually, I'll just try concatenating the 5 channels. Thanks anyways.

[Victor Bazarov]

Mark wrote:

On May 14, 1:14 pm, Mark <[e-mail]:

>On May 14, 1:07 pm, Victor Bazarov <[e-mail]:

>>Mark wrote:
I'm using AdaBoost on optical flow data to learn to recognize human
actions. I've split my optical flow data into 5 channels (positive x,
negative x, positive y, negative y, and 'zero'). I know how to use
AdaBoost on a single channel, but I'm not sure how to combine the 5
channels to form a better classifier. Do I run AdaBoost on each
channel, and then pass the 5 outputs to another AdaBoost, or should I
simply concatenate the 5 channels and run AdaBoost once on all of
them?
What's "AdaBoost"?
V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask

AdaBoost is a boosting/machine learning algorithm. It's short for
Adaptive Boosting (http://en.wikipedia.org/wiki/AdaBoost). My
question isn't exactly specific to AdaBoost. The output of AdaBoost
is a class. For example, I've got some videos where I am waving my
hand, and some where I am not. AdaBoost spits out "pos" for waving,
or "neg" for not waving. I could run AdaBoost on my 5 channels, and
do something like a majority vote, or use AdaBoost to figure out how
many "pos"s it actually needs.... or I can perhaps pass the confusion
matrix in as input... Basically, I need to figure out how to make it
multi-layer. (Btw, I'm achieving a 13% error rate using just 1
channel! This is exciting)

Mark

Actually, I'll just try concatenating the 5 channels. Thanks anyways.

You're welcome; glad to have helped... I am still perplexed about the
reasons that led you to the *C++ language* newsgroup.

V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask

[Mark]

On May 14, 3:30 pm, Victor Bazarov <[e-mail]:

Mark wrote:

On May 14, 1:14 pm, Mark <[e-mail]:

On May 14, 1:07 pm, Victor Bazarov <[e-mail]:

>Mark wrote:
I'm using AdaBoost on optical flow data to learn to recognize human
actions. I've split my optical flow data into 5 channels (positive x,
negative x, positive y, negative y, and 'zero'). I know how to use
AdaBoost on a single channel, but I'm not sure how to combine the 5
channels to form a better classifier. Do I run AdaBoost on each
channel, and then pass the 5 outputs to another AdaBoost, or should I
simply concatenate the 5 channels and run AdaBoost once on all of
them?
What's "AdaBoost"?
V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask
AdaBoost is a boosting/machine learning algorithm. It's short for
Adaptive Boosting (http://en.wikipedia.org/wiki/AdaBoost). My
question isn't exactly specific to AdaBoost. The output of AdaBoost
is a class. For example, I've got some videos where I am waving my
hand, and some where I am not. AdaBoost spits out "pos" for waving,
or "neg" for not waving. I could run AdaBoost on my 5 channels, and
do something like a majority vote, or use AdaBoost to figure out how
many "pos"s it actually needs.... or I can perhaps pass the confusion
matrix in as input... Basically, I need to figure out how to make it
multi-layer. (Btw, I'm achieving a 13% error rate using just 1
channel! This is exciting)

Mark

Actually, I'll just try concatenating the 5 channels. Thanks anyways.

You're welcome; glad to have helped... I am still perplexed about the
reasons that led you to the *C++ language* newsgroup.

V
--
Please remove capital 'A's when replying by e-mail
I do not respond to top-posted replies, please don't ask

I guess it's not really specific to C++ either, huh? I'm using a C++
implementation of AdaBoost (called 'MultiBoost')... nevertheless, I
didn't think there was a group dedicated to boosting, although perhaps
I should have searched for machine learning or something.

Mark