python - Theano: Fastest large-scale symmetric linear system solve -
does know fastest method solving large-scale symmetric linear systems in theano (i.e. solve ax=b x a symmetric)?
in specific case, trying solve several linear systems a symmetric, dense, , guaranteed positive definite. presently, using theano.tensor.slinalg.solve op not appear include options exploit specific matrix structures scipy analogue. noticed theano.tensor.slinalg.solve (note capitalization) has options different matrix structures including symmetric and symmetric positive definite, not sure how use because not take a or b arguments.
the "solve" , "solve" ops found here.
edit: figured out how use "solve". can used follows:
import theano import theano.tensor t theano.tensor.slinalg import solve import numpy np b = t.vector('b') = t.matrix('a') sym_lin_solve = solve(a_structure='symmetric') x = sym_lin_solve(a, b) sym_lin_solve_func = theano.function(inputs=[a, b], outputs=x) # answer should array([-1., 3.]) print sym_lin_solve_func(np.array([[1., 2.],[2., 1.]]), np.array([5., 1.])) however, seems a_structure='symmetric' not anything.
Comments
Post a Comment