python - Operations on huge dense matrices in numpy -
for purpose of training neural network, @ point have huge 212,243 × 2500 dense matrix phi
, , vectors y
(212243) , w
(2500), stored numpy
arrays of doubles. i'm trying compute is
w = dot(pinv(phi), y) # serialize w... r = dot(w, transpose(phi)) # serialize r...
my machine has 6 gb of ram , 16 gb of swap on ubuntu x64. started computation twice , twice has ended system (not python) swap errors after hour of work.
is there way perform computation on computer? doesn't need done python.
if don't need pseudoinverse else computing w
, replace line with:
w = np.linalg.lstsq(phi, y)[0]
on system runs 2x faster, , uses half intermediate storage.
Comments
Post a Comment