def parallel_grid_search(settings, folds, data, features, estimate_g):
n_tests = settings.grid_tests
n_proc = settings.number_proc
##starts the parameters
exps_c = [-7, -5, -3, -1, 1, 3, 5, 7, 9, 11, 13, 15]
if estimate_g: ##in case we have to estimate g we define a range
exps_g = [-13, -11, -9, -7, -5, -3, -1, 1, 3, 5, 7]
else: ##in case we dont just define it already
exps_g = [1]
params_c = []
params_g = []
for e in exps_c:
params_c.append(2**e)
for e in exps_g:
params_g.append(2**e)
param_grid = [(x, y) for x in params_c for y in params_g]
com = mp.Pipe()
shared_lock = mp.Lock()
larger_search = True ##first we will search with huge intervals as recomended on the paper then when a larger interval is found we will search on lower intervals inside that interval
while (len(param_grid) != 1): ##while we dont have one set
##generates n random subsets
subsets = utils.generate_random_sets(features, n_tests, 3,
len(features))
division_subsets = utils.divide_list(n_proc, subsets)
##parallelize work to it becomes faster
workers = []
for i in range(1, n_proc):
p = mp.Process(target=grid_search,
args=(i, division_subsets[i], folds, data,
param_grid, shared_lock, com,
settings.svm_kernel))
workers.append(p)
p.start()
grid_search(0, division_subsets[0], folds, data, param_grid,
shared_lock, com, settings.svm_kernel)
for w in workers: ##hopefully this waits for every process to finish...
w.join()
output, input = com
best_params = output.recv()
if len(
best_params
) != n_tests: ##either we received the score of every feature or some error happened
print "Didn't receive best param for every subset. ERROR"
quit()
param_grid, larger_search = get_top_selected(best_params,
larger_search, estimate_g)
##returns c and gamma
return param_grid[0][0], param_grid[0][1]
def parallel_grid_search(settings, folds, data, features, estimate_g):
n_tests = settings.grid_tests
n_proc = settings.number_proc
##starts the parameters
exps_c = [-7, -5, -3, -1, 1, 3, 5, 7, 9, 11, 13, 15]
if estimate_g: ##in case we have to estimate g we define a range
exps_g = [-13, -11, -9, -7, -5, -3, -1, 1, 3, 5, 7]
else: ##in case we dont just define it already
exps_g = [1]
params_c = []
params_g = []
for e in exps_c:
params_c.append(2**e)
for e in exps_g:
params_g.append(2**e)
param_grid = [(x, y) for x in params_c for y in params_g]
com = mp.Pipe()
shared_lock = mp.Lock()
larger_search = True ##first we will search with huge intervals as recomended on the paper then when a larger interval is found we will search on lower intervals inside that interval
while (len(param_grid) != 1): ##while we dont have one set
##generates n random subsets
subsets = utils.generate_random_sets(features, n_tests, 3, len(features))
division_subsets = utils.divide_list(n_proc, subsets)
##parallelize work to it becomes faster
workers = []
for i in range(1, n_proc):
p = mp.Process(target=grid_search, args=(i, division_subsets[i], folds, data, param_grid, shared_lock, com, settings.svm_kernel))
workers.append(p)
p.start()
grid_search(0, division_subsets[0], folds, data, param_grid, shared_lock, com, settings.svm_kernel)
for w in workers: ##hopefully this waits for every process to finish...
w.join()
output, input = com
best_params = output.recv()
if len(best_params) != n_tests: ##either we received the score of every feature or some error happened
print "Didn't receive best param for every subset. ERROR"
quit()
param_grid, larger_search = get_top_selected(best_params, larger_search, estimate_g)
##returns c and gamma
return param_grid[0][0], param_grid[0][1]
def get_improve_prob(features, n_tests, dataset, folds, n_proc, p_features, shared_lock, com, cost, gamma, svm_kernel):
test_sets = utils.generate_random_sets(features, n_tests, 3, 10)
accs = []
for test in test_sets: ##test each dataset agaisnt the classifier
acc = classification_part.classify(folds, dataset, test, cost, gamma, svm_kernel)
accs.append(acc)
##split tests among processes
#print "accuracies subsets:", acc
workers = []
for i in range(1, n_proc):
p = mp.Process(target=calculate_improve_prob, args=(i, p_features[i], dataset, test_sets, accs, folds, shared_lock, com, cost, gamma, svm_kernel))
workers.append(p)
p.start()
calculate_improve_prob(0, p_features[0], dataset, test_sets, accs, folds, shared_lock, com, cost, gamma, svm_kernel)
for w in workers:
w.join()
return
def get_improve_prob(features, n_tests, dataset, folds, n_proc, p_features, shared_lock, com, cost, gamma, svm_kernel):
test_sets = utils.generate_random_sets(features, n_tests, 3, 10)
accs = []
for test in test_sets: ##test each dataset agaisnt the classifier
acc = classification_part.classify(folds, dataset, test, cost, gamma, svm_kernel)
accs.append(acc)
##split tests among processes
# print "accuracies subsets:", acc
workers = []
for i in range(1, n_proc):
p = mp.Process(
target=calculate_improve_prob,
args=(i, p_features[i], dataset, test_sets, accs, folds, shared_lock, com, cost, gamma, svm_kernel),
)
workers.append(p)
p.start()
calculate_improve_prob(0, p_features[0], dataset, test_sets, accs, folds, shared_lock, com, cost, gamma, svm_kernel)
for w in workers:
w.join()
return
