def run_DANE_ridgeregression_experiment_2(N, m, max_iter, flag, data, w_opt,
mode, max_inner_iter,
sampling_flag_rate, rate_param):
'''we give 0 for data and w_opt if we want to draw them fresh, but
give them as input if we want to use the same ones and run on different number of machines or different iteration numbers'''
print m
# Hmm
# setting the objective and DANE parameters:
objective_param = 0.005
eta = 1.0
mu = 0.000001
# mu = 0
DANE_params = [eta, mu]
if flag == 0:
# generating N 500-d points from y = <x, w_opt> + noise:
w_opt = np.ones([500, 1]) # line parameters
# distribution for data points:
mean = np.zeros([500])
cov = np.diag((np.array(range(1, 501)))**(-1.2)) # ** (-1.2)
# draw random data points:
X = np.random.multivariate_normal(mean, cov, (N))
# estimate y for x given w:
Y = np.dot(X, w_opt)
# adding the noise :
noise = np.array(np.random.standard_normal(size=(N, 1)))
Y = Y + noise
data = np.concatenate((X, Y), axis=1)
w_opt = np.reshape(w_opt, (500)) # this might be not needed anymore
# '''better to change it to use the machines rather than directly using ridge-regression class
# since we want to have it in general form'''
# mainrg = Ridge_regression( X, np.reshape(Y, (N)), [0.005] )
# main_opt_eval = mainrg.eval(w_opt)
# print 'first main_opt_eval, ', main_opt_eval
# I am calling initialize_machines to set up our computing machines:
machines = initialize_machines(m, data)
'''Running Dane procedure:'''
evals, runtimes, w_ans, number_of_gradients, number_of_gradients_2 = DANE_procedure(
machines, w_opt, 'ridge_regression', objective_param, max_iter,
DANE_params[0], DANE_params[1], mode, max_inner_iter,
sampling_flag_rate, rate_param)
return evals, runtimes, w_ans, w_opt, data, number_of_gradients, number_of_gradients_2
def run_DANE_ridgeregression_experiment_2(N, m, max_iter, flag, data, w_opt , mode , max_inner_iter , sampling_flag_rate , rate_param): '''we give 0 for data and w_opt if we want to draw them fresh, but give them as input if we want to use the same ones and run on different number of machines or different iteration numbers''' print m # Hmm # setting the objective and DANE parameters: objective_param = 0.005 eta=1.0 mu=0.000001 # mu = 0 DANE_params = [ eta , mu ] if flag ==0: # generating N 500-d points from y = <x, w_opt> + noise: w_opt = np.ones( [ 500, 1 ] ) # line parameters # distribution for data points: mean = np.zeros( [ 500 ] ) cov = np.diag( (np.array(range(1, 501))) ** ( -1.2 ) ) # ** (-1.2) # draw random data points: X = np.random.multivariate_normal(mean, cov, ( N )) # estimate y for x given w: Y = np.dot( X , w_opt ) # adding the noise : noise = np.array(np.random.standard_normal( size=( N, 1) )) Y = Y + noise data = np.concatenate(( X , Y ), axis = 1 ) w_opt = np.reshape(w_opt, (500)) # this might be not needed anymore # '''better to change it to use the machines rather than directly using ridge-regression class # since we want to have it in general form''' # mainrg = Ridge_regression( X, np.reshape(Y, (N)), [0.005] ) # main_opt_eval = mainrg.eval(w_opt) # print 'first main_opt_eval, ', main_opt_eval # I am calling initialize_machines to set up our computing machines: machines = initialize_machines( m, data ) '''Running Dane procedure:''' evals, runtimes, w_ans , number_of_gradients , number_of_gradients_2 = DANE_procedure( machines , w_opt, 'ridge_regression', objective_param , max_iter, DANE_params[0] , DANE_params[1] , mode , max_inner_iter , sampling_flag_rate , rate_param ) return evals, runtimes, w_ans , w_opt, data , number_of_gradients , number_of_gradients_2