Question

我想像论文https://pdfs.semanticscholar.org/4957/0fb292730a435b1e9e3a85a5dac36f784821.pdf中那样用CVXPY解决GROUP DANTZIG SELECTOR 但这给了我以下错误：

Traceback (most recent call last):   File "/home/bognev/PycharmProjects/BSBL-LSTM-Net/dantzig_selector.py", line 83, in <module>
    constr_f = [cp.norm(d_f,"Inf") <= p, sum(a) <= q]   File "/home/bognev/anaconda3/lib/python3.7/site-packages/cvxpy/atoms/norm.py", line 43, in norm
    x = Expression.cast_to_const(x)   File "/home/bognev/anaconda3/lib/python3.7/site-packages/cvxpy/expressions/expression.py", line 417, in cast_to_const
    return expr if isinstance(expr, Expression) else cvxtypes.constant()(expr)   File "/home/bognev/anaconda3/lib/python3.7/site-packages/cvxpy/expressions/constants/constant.py", line 44, in __init__
    self._value = intf.DEFAULT_INTF.const_to_matrix(value)   File "/home/bognev/anaconda3/lib/python3.7/site-packages/cvxpy/interface/numpy_interface/ndarray_interface.py", line 50, in const_to_matrix
    return result.astype(numpy.float64) ValueError: setting an array element with a sequence.

问题是我想找到最大值（EXPRESSIONS），而cvxpy无法解决它。该公式在MATLAB CVX中工作正常也许有人知道如何避免此错误？

lambdas_f = cp.Parameter(nonneg=True)
lambdas_f.value = 30
# Define problem
x_v_f = cp.Variable(input_size*num_groups)
p_f = cp.Variable(1)
q_f = cp.Variable(1)

G = np.zeros((input_size,num_groups,output_size))
a_f = []
d_f = []

for ii in range(input_size):
    G[ii,:,:] = (F@A[:,ii:input_size*num_groups:input_size]).transpose()
for ii in range(input_size):
    a_f.append(cp.norm(x_v_f[ii:input_size*num_groups:input_size],2))
for ii in range(input_size):
    d_f.append(cp.norm(G[ii]@(y-F@A@x_v_f),2))
objective_f = 0.5*p**2+lambdas*q
constr_f = [cp.norm(d_f,"Inf") <= p, sum(a) <= q]
prob_f = cp.Problem(cp.Minimize(objective_f), constr_f)
prob_f.solve()

如何在CVXPY中正确制定GROUP DANTZIG SELECTOR

0 个答案: