def test_mae():
w0 = np.array([0.5, 0.125])
w1 = np.array([0.125, (1 - 0.5 - 2 * 0.125)])
d1 = np.array([0, 0])
d2 = np.array([0, 1])
d3 = np.array([1, 0])
d4 = np.array([1, 1])
nt.assert_almost_equal(cl.mae_from_distribution(d1, w0, w1), w1.sum())
nt.assert_almost_equal(cl.mae_from_distribution(d2, w0, w1), w1[0] + w0[1])
nt.assert_almost_equal(cl.mae_from_distribution(d3, w0, w1), w0[0] + w1[1])
nt.assert_almost_equal(cl.mae_from_distribution(d4, w0, w1), w0.sum())
def main(xpl_data, num_features=None, num_iterations=None, save_todir=None):
data = xpl_data.data
# copies frequency data as original frequencies are used towards the end to estimate training error
w0 = xpl_data.freq0.copy()
w1 = xpl_data.freq1.copy()
error_list = []
mae_list = []
DEC = np.zeros(w0.shape)
GVector = []
i=0
winfile = "window_"
win = ".win"
png = ".png"
w0_train, w1_train = cl.normalize_table(xpl_data.freq0, xpl_data.freq1)
file = open(save_todir+"Error.txt", "w")
for i in range(num_iterations):
indices, feature_list, _ = ft.cmim(data, w0, w1, num_features)
tw.to_window_file(indices, xpl_data.winshape, save_todir+winfile+str(i)+win)
tw.to_image_file(indices,xpl_data.winshape, save_todir+winfile+str(i)+png, scale=8)
w0, w1 = cl.normalize_table(w0, w1)
w0, w1, updated_decision, cls_error = cl.apply_feature_selection(data, indices, w0, w1)
unique_array, unique_index = cl._apply_projection(data, indices)
xplutil.write_minterm_file(save_todir+"mtm_"+str(i),indices, xpl_data.winshape,unique_array,updated_decision[unique_index])
str_to_file = "Classification error for iteration " + str(i) +" = "+ str(cls_error) +".\n"
file.write(str_to_file)
error_list.append(cls_error)
bt = cl.beta_factor(cls_error)
gam = np.log(1/bt)
GVector = np.append(GVector,gam)
#DEC represents the Decision Table. Each column represents the decision
#for an iteration
DEC = np.column_stack((DEC,updated_decision))
aux_dec = DEC
aux_dec = np.delete(aux_dec,0, axis=1)
hypothesis = cl.adaboost_decision(aux_dec, GVector)
MAE_t = cl.mae_from_distribution(hypothesis,w0_train, w1_train)
mae_list = np.append(mae_list,MAE_t)
str_to_file = "MAE for iteration " + str(i) +" = "+ str(MAE_t) +".\n\n"
file.write(str_to_file)
#Must delete the first column because it contains only Zeros as it was initialized with np.zeros()
DEC = np.delete(DEC,0, axis=1)
hypothesis = cl.adaboost_decision(DEC, GVector)
MAE = cl.mae_from_distribution(hypothesis, w0_train, w1_train)
str_to_file = "Final MAE = "+ str(MAE)
file.write(str_to_file)
#print MAE
file.close()
gra.plot_MAE_iter(np.array(range(num_iterations)), np.array(mae_list))
def train(xpl_data, n_features, n_iterations, dirpath):
Xdata = xpl_data.data
win = xpl_data.windata
# copies frequency data as original frequencies are used towards the end to estimate training error
w0 = xpl_data.freq0.copy()
w1 = xpl_data.freq1.copy()
error_list = []
mae_list = []
GVector = []
DEC = np.zeros(w0.shape)
total = float(np.sum([w0, w1]))
w0_train = w0/total
w1_train = w1/total
file = open(dirpath+"MAE_training.txt", "w")
for i in range(n_iterations):
indices, feature_list, _ = ft.cmim(Xdata, w0, w1, n_features)
indices = np.sort(indices)
triosw.to_window_file(indices, xpl_data.winshape, dirpath+"window_"+str(i)+".win")
triosw.to_image_file(indices,xpl_data.winshape, dirpath+"window_"+str(i)+".png", scale=8)
total = float(np.sum([w0, w1]))
w0 = w0/total
w1 = w1/total
w0, w1, updated_decision, cls_error = clf.apply_feature_selection(Xdata, indices, w0, w1)
unique_array, unique_index = clf._apply_projection(Xdata, indices)
xplutil.write_minterm_file(dirpath+"mtm"+str(i),indices, xpl_data.winshape, unique_array,updated_decision[unique_index])
#str_to_file = "Classification error for iteration " + str(i) +" = "+ str(cls_error) +".\n"
#file.write(str_to_file)
error_list.append(cls_error)
bt = clf.beta_factor(cls_error)
gam = np.log(1/bt)
GVector = np.append(GVector,gam)
#DEC represents the Decision Table. Each column represents the decision for an iteration
DEC = np.column_stack((DEC,updated_decision))
aux_dec = DEC
aux_dec = np.delete(aux_dec,0, axis=1)
hypothesis = clf.adaboost_decision(aux_dec, GVector)
MAE_t = clf.mae_from_distribution(hypothesis,w0_train, w1_train)
mae_list = np.append(mae_list,MAE_t)
str_to_file = str(i) +", "+ str(MAE_t) +"\n"
file.write(str_to_file)
#Must delete the first column because it contains only Zeros as it was initialized with np.zeros()
DEC = np.delete(DEC,0, axis=1)
hypothesis = clf.adaboost_decision(DEC, GVector)
#MAE = clf.mae_from_distribution(hypothesis,w0_train, w1_train)
#str_to_file = "Final MAE = "+str(MAE)
#file.write(str(MAE))
file.close()
plot_MAE(np.array(range(n_iterations)), np.array(mae_list), dirpath)
return Ensemble(xpl_data, win, n_features, n_iterations, error_list, mae_list,dirpath)
