def candidate_test(names, num_nodes, alphas, betas, dict_partition,
dim_partition, X, D, D_mean):
plt_accs, plt_ops = {}, {}
for name, node in zip(names, num_nodes):
plt_ops[name] = []
plt_accs[name] = []
for alpha, beta in zip(alphas, betas):
nnuf = NNUForest(alpha=alpha, beta=beta, num_nodes=node,
num_dims=dim, dim_partition=dim_partition,
dict_partition=dict_partition)
nnuf.build_index(D)
pct, avg_candidates = pct_found(nnuf, X, D, D_mean)
plt_accs[name].append(pct)
plt_ops[name].append(num_kops(avg_candidates, len(X[0][0])))
return plt_ops, plt_accs
def accuracy_test(names, num_l1_nodes, alphas, betas, dict_partition, dim_partition,
X, Y, D, D_mean, sss, model='NNUF', num_l2_nodes=0):
plt_accs, plt_ops = {}, {}
plt_ops = []
plt_accs = []
for alpha, beta in zip(alphas, betas):
print alpha, beta
if model == 'NNUF':
nnuf = NNUForest(alpha=alpha, beta=beta, num_nodes=num_l1_nodes, num_dims=dim,
dim_partition=dim_partition,
dict_partition=dict_partition)
elif model == 'DictDim':
nnuf = NNUDictDim(num_l1_nodes, num_l2_nodes, alpha, beta)
nnuf.build_index(D)
ret_tup = Parallel(n_jobs=8, verbose=0)(delayed(classify)(nnuf, X_window, Y,
train_index, test_index)
for train_index, test_index in sss)
accs, num_candidates = zip(*ret_tup)
avg_candidates = np.sum(num_candidates) / float(num_X*num_folds)
plt_accs.append(np.mean(accs))
plt_ops.append(num_kops(avg_candidates, len(X[0][0])))
return plt_ops, plt_accs
