tf.raw_ops.SparseApplyProximalGradientDescent

robots: noindex

Sparse update '*var' as FOBOS algorithm with fixed learning rate.