SGMM takes quite long long time for training when the input features are about 400,
particularly for the 'sgmm2-acc-stats' process...
is it something wrong or it takes longer with high dimension features?
or I should fine-tune some parameters to run it properly/successfully with such high dimension?
e.g. currently I still use:
num_gauss=400 num_leaves=5000 num_substates=8000
thanks in advance!
Fei
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
It is expected to be slow for high-dimensional features because some parts
of the training take linear time in the dimension and some quadratic (e.g.
with covariance-related things).
Dan
SGMM takes quite long long time for training when the input features are
about 400,
particularly for the 'sgmm2-acc-stats' process...
is it something wrong or it takes longer with high dimension features?
or I should fine-tune some parameters to run it properly/successfully with
such high dimension?
e.g. currently I still use:
num_gauss=400 num_leaves=5000 num_substates=8000
Hi Kaldis,
SGMM takes quite long long time for training when the input features are about 400,
particularly for the 'sgmm2-acc-stats' process...
is it something wrong or it takes longer with high dimension features?
or I should fine-tune some parameters to run it properly/successfully with such high dimension?
e.g. currently I still use:
num_gauss=400 num_leaves=5000 num_substates=8000
thanks in advance!
Fei
It is expected to be slow for high-dimensional features because some parts
of the training take linear time in the dimension and some quadratic (e.g.
with covariance-related things).
Dan
On Thu, Jan 29, 2015 at 2:13 PM, Fei Xiong xffmqjx@users.sf.net wrote: