diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/algorithms.py | 16 | ||||
| -rw-r--r-- | src/convex_optimization.py | 48 | ||||
| -rw-r--r-- | src/make_plots.py | 10 |
3 files changed, 37 insertions, 37 deletions
diff --git a/src/algorithms.py b/src/algorithms.py index 926d9bc..9ebccf1 100644 --- a/src/algorithms.py +++ b/src/algorithms.py @@ -31,7 +31,7 @@ def greedy_prediction(G, cascades): return G_hat -def recovery_l1obj_l2constraint(G, cascades, floor_cstt, passed_function, +def recovery_passed_function(G, cascades, floor_cstt, passed_function, *args, **kwargs): """ Returns estimated graph from convex program specified by passed_function @@ -40,11 +40,13 @@ def recovery_l1obj_l2constraint(G, cascades, floor_cstt, passed_function, G_hat = cascade_creation.InfluenceGraph(max_proba=None) G_hat.add_nodes_from(G.nodes()) + f_x, f_xz = passed_function(*args, **kwargs) + for node in G_hat.nodes(): print(node) try: M, w = cascade_creation.icc_matrixvector_for_node(cascades, node) - p_node, __ = passed_function(M,w, *args, **kwargs) + p_node, _ = convex_optimization.diff_and_opt(M, w, f_x, f_xz) G_hat = cascade_creation.add_edges_from_proba_vector(G=G_hat, p_node=p_node, node=node, floor_cstt=floor_cstt) except timeout.TimeoutError: @@ -86,15 +88,15 @@ def test(): unit test """ G = cascade_creation.InfluenceGraph(max_proba = .8) - G.erdos_init(n = 10, p = .2) + G.erdos_init(n = 100, p = .2) import time t0 = time.time() A = cascade_creation.generate_cascades(G, .2, 1000) - if 1: - G_hat = greedy_prediction(G, A) if 0: - G_hat = recovery_l1obj_l2constraint(G, A, - passed_function=convex_optimization.l1obj_l2penalization, + G_hat = greedy_prediction(G, A) + if 1: + G_hat = recovery_passed_function(G, A, + passed_function=convex_optimization.type_lasso, floor_cstt=.1, lbda=10) correctness_measure(G, G_hat, print_values=True) diff --git a/src/convex_optimization.py b/src/convex_optimization.py index 1f34508..163b6d5 100644 --- a/src/convex_optimization.py +++ b/src/convex_optimization.py @@ -44,55 +44,52 @@ def sparse_recovery(M_val, w_val, lbda): return diff_and_opt(theta, theta_, M, M_val, w, lbda, y) -@timeout.timeout(20) -def type_lasso(M_val, w_val, lbda): +def type_lasso(lbda): """ Solves: min - sum_j theta_j + lbda*|e^{M*theta} - (1 - w)|_2 s.t theta_j <= 0 """ - assert len(M_val) == len(w_val) - - if M_val.dtype == bool: - M_val = M_val.astype('float32') - if type(lbda) == int or type(lbda) == float: lbda = np.array(lbda) + lbda = theano.shared(lbda.astype(theano.config.floatX)) + theta = tensor.row().T theta_ = theta.flatten() - M = theano.shared(M_val.astype(theano.config.floatX)) - w = theano.shared(w_val.astype(theano.config.floatX)) - lbda = theano.shared(lbda.astype(theano.config.floatX)) + M = tensor.matrix(name='M', dtype=theano.config.floatX) + w = tensor.vector(name='w', dtype=theano.config.floatX) y = lbda * (theta_).norm(1) + ( tensor.exp(M.dot(theta_)) - (1 - w)).norm(2) - return diff_and_opt(theta, theta_, M, M_val, w, lbda, y) - - -def diff_and_opt(theta, theta_, M, M_val, w, lbda, y): z = tensor.row().T z_ = z.flatten() - m, n = M_val.shape - y_diff = tensor.grad(y, theta_) y_hess = z[0] * theano.gradient.hessian(y, theta_) - f_x = theano.function([theta], [y, y_diff], allow_input_downcast=True) - f_xz = theano.function([theta, z], [y, y_diff, y_hess], - allow_input_downcast=True) + f_x = theano.function([theta, M, w], [y, y_diff], + allow_input_downcast=True) + f_xz = theano.function([theta, z, M, w], [y, y_diff, y_hess], + allow_input_downcast=True) + + return f_x, f_xz + +@timeout.timeout(8) +def diff_and_opt(M_val, w_val, f_x, f_xz): + + m, n = M_val.shape def F(x=None, z=None): if x is None: return 0, cvxopt.matrix(-.001, (n,1)) elif z is None: - y, y_diff = f_x(x) + y, y_diff = f_x(x, M_val, w_val) return cvxopt.matrix(float(y), (1, 1)),\ cvxopt.matrix(y_diff.astype("float64")).T else: - y, y_diff, y_hess = f_xz(x, z) + y, y_diff, y_hess = f_xz(x, z, M_val, w_val) return cvxopt.matrix(float(y), (1, 1)), \ cvxopt.matrix(y_diff.astype("float64")).T, \ cvxopt.matrix(y_hess.astype("float64")) @@ -104,7 +101,7 @@ def diff_and_opt(theta, theta_, M, M_val, w, lbda, y): #cvxopt.solvers.options['feastol'] = 2e-5 #cvxopt.solvers.options['abstol'] = 2e-5 #cvxopt.solvers.options['maxiters'] = 100 - cvxopt.solvers.options['show_progress'] = False + cvxopt.solvers.options['show_progress'] = True try: theta = cvxopt.solvers.cp(F, G, h)['x'] except ArithmeticError: @@ -128,13 +125,14 @@ def test(): M_val, w_val = cascade_creation.icc_matrixvector_for_node(A, 0) #Type lasso - if 0: - p_vec, theta = type_lasso(M_val, w_val, lbda) + if 1: + f_x, f_xz = type_lasso(lbda) + p_vec, _ = diff_and_opt(M_val, w_val, f_x, f_xz) print(p_vec) print(G.mat) #Sparse recovery - if 1: + if 0: p_vec, theta = sparse_recovery(M_val, w_val, lbda) print(p_vec) print(G.mat) diff --git a/src/make_plots.py b/src/make_plots.py index 3ac8362..c1cafe7 100644 --- a/src/make_plots.py +++ b/src/make_plots.py @@ -23,7 +23,7 @@ def compare_greedy_and_lagrange_cs284r(): #Lagrange Objective G_hat = algorithms.recovery_l1obj_l2constraint(G, A, passed_function=convex_optimization.type_lasso, - floor_cstt=.1, lbda=10) + floor_cstt=.05, lbda=10) algorithms.correctness_measure(G, G_hat, print_values=True) @@ -32,11 +32,11 @@ def test(): unit test """ G = cascade_creation.InfluenceGraph(max_proba=.8) - G.erdos_init(n=20, p=.2) + G.erdos_init(n=30, p=.2) A = cascade_creation.generate_cascades(G, p_init=.1, n_cascades=1000) - G_hat = algorithms.recovery_l1obj_l2constraint(G, A, - passed_function=convex_optimization.sparse_recovery, - floor_cstt=.1, lbda=1) + G_hat = algorithms.recovery_passed_function(G, A, + passed_function=convex_optimization.type_lasso, + floor_cstt=.001, lbda=.0001) algorithms.correctness_measure(G, G_hat, print_values=True) if __name__=="__main__": |