fix projected GD

examples/reconstruct/test-projected-gd.jl

@@ -10,28 +10,6 @@ X = h5read("X.h5", "X")
 A = h5read("A.h5", "A")
 Y = h5read("Y.h5", "Y")
 
-
-# A_array = [A[i,:] for i in 1:size(A,1)]
-
-# n, x = recovery_theta1(size(X), A_array, Y);
-# x = reshape(x, size(X))
-# relative_error(X,x)
-
-# hosvd_x = hosvd(x);
-
-# core = hosvd_x["core"]
-# multirank(core)
-
-# core
-
-# core[abs.(core).<1e-5] .= 0
-# multirank(core)
-
-# hosvd_x["core"] = core
-# x_re = hosvd_reconstruct(hosvd_x)
-# @assert relative_error(X,x_re) < 1E-2
-
-
 function projection(X, rank)
   hosvd_x = hosvd(X);
   core_tmp = hosvd_x["core"]
@@ -64,18 +42,3 @@ rank = 2
 res = optimal_gradient_descent(ones(size(X)), 1, 5000, rank);
 relative_error(X, res.x)
 plot(res.hist, yaxis=:log, labels="resnorm")
-
-# size_tensor = (3,3,3) # size of the tensor
-# r = 1 # rank of the tensor
-# coefs = randn(r) # coefficients of rank-1 tensors
-# x0 = create_tensor_CP(size_tensor, coefs, "normal"); # create a rank-1 tensor by gaussian distribution
-# m = 3^6-1 # number of measurements
-# A, b = create_measurements(x0, m); # create gaussian linear measurements
-# # A should be a list of length m, each elements is a flattened tensor, i.e. a vector indeed of size 4*4*4=64.
-# # b is also a list of length m, each element is the number b_i = <A_i, x0>.
-
-# n, x = recovery_theta1(size_tensor, A, b); # using nuclear-2 theta-1 norm minimization for recovery
-# # n is the theta-1 norm, x is the recovered tensor
-
-# x = reshape(x, size(x0))# reshape the recovered tensor to its original shape
-# relative_error(x0,x) # compare the difference