def __init__(self, parent, title, dir):
ffp = FaceFileParser()
ffp.add_dir(dir)
data_imgs = ffp.get_data()
wx.Frame.__init__(self, parent, title=title, size=(800, 650))
self.imgs = [[], [], [], []]
self.images = [[], [], [], []]
for img in data_imgs:
self.imgs[img[0]].append(img)
for a in range(0, 4):
self.images[a] = self.make_images(self.imgs[a])
for a in range(0, 4):
for b in range(0, len(self.images[a])):
i = self.images[a][b].Scale(180, 36)
i = i.ConvertToBitmap()
wx.StaticBitmap(self, -1, i, (5 + (190 * a), 5 + (46 * b)))
#i = self.images[0].Scale(240,80,wx.IMAGE_QUALITY_NORMAL)
#self.show_img = wx.StaticBitmap(self,-1,i.ConvertToBitmap(),(5,5))
#self.img_c = 0
#self.next_b = wx.Button(self,label='Next',pos=(5,90))
#self.Bind(wx.EVT_BUTTON,self.next_img,self.next_b)
self.Show()
def __init__(self,src_dir,dest_dir):
ffp = FaceFileParser()
ffp.add_dir(src_dir)
data_imgs = ffp.get_data()
self.img_lists = [[],[],[],[]]
self.dest_dir = dest_dir
for img in data_imgs:
self.img_lists[img[0]].append(list(img[1]))
def train(self, event):
if self.train_f_name_tb.GetValue() != '':
ffp = FaceFileParser()
ffp.add_dir(self.train_f_name_tb.GetValue())
datalist = ffp.get_data()
layers = int(self.train_numLayers_tb.GetValue())
nodes = int(self.train_numNodes_tb.GetValue())
epochs = int(self.train_numEpochs_tb.GetValue())
nettype = self.netbox.GetSelection()
print layers
print nodes
print epochs
print nettype
if nettype == 0:
fc = SupervisedFacialClassifier()
if layers == 1:
errorData = fc.alternateTrain(datalist, [(nodes)], epochs,
5)
elif layers == 2:
errorData = fc.alternateTrain(datalist, (nodes, nodes),
epochs, 5)
elif layers == 3:
errorData = fc.alternateTrain(datalist,
(nodes, nodes, nodes),
epochs, 5)
print 'ErrorData:\n'
print errorData
print '\n'
for entry in errorData['training_error']:
print entry
else:
fc = UnsupervisedFacialClassifier(2, 2)
errorData = fc.train(datalist)
def train(self, event):
if self.train_f_name_tb.GetValue() != '':
ffp = FaceFileParser()
ffp.add_dir(self.train_f_name_tb.GetValue())
datalist = ffp.get_data()
layers = int(self.train_numLayers_tb.GetValue())
nodes = int(self.train_numNodes_tb.GetValue())
epochs = int(self.train_numEpochs_tb.GetValue())
nettype = self.netbox.GetSelection()
print layers
print nodes
print epochs
print nettype
if nettype == 0:
fc = SupervisedFacialClassifier()
if layers == 1:
errorData = fc.alternateTrain(datalist, [(nodes)], epochs, 5)
elif layers == 2:
errorData = fc.alternateTrain(datalist, (nodes,nodes), epochs, 5)
elif layers == 3:
errorData = fc.alternateTrain(datalist, (nodes,nodes,nodes), epochs, 5)
print 'ErrorData:\n'
print errorData
print '\n'
for entry in errorData['training_error']:
print entry
else:
fc = UnsupervisedFacialClassifier(2,2)
errorData = fc.train(datalist)
for data_dir in os.walk(root_data_dir):
(dirname, subdirs, files) = data_dir
dirs_to_test.append(dirname)
# Write the trial data out to a file
outfile = open(outfile_name, 'w')
outfile.write(
"DIRNAME, LEARNINGRATE, DECAY, EPOCHS, TRAINING_ERR, TRAINING_ERR_STD, TESTING_ERR, TESTING_ERR_STD\n"
)
for dirname in dirs_to_test:
for learningrate in learningrates_to_test:
for neighbourdecay in neighbourdecays_to_test:
for epochs in epochs_to_test:
ffp = FaceFileParser(verbose=False)
ffp.add_dir(dirname)
datalist = ffp.get_data()
print "################################"
print " %s, learningrate: %s, decay: %s, epochs: %s" %\
(dirname, learningrate, neighbourdecay, epochs)
trial_avg_testing_err = []
trial_avg_training_err = []
# Run trials
for trial in range(num_trials):
print "Trial #%i" % trial
fc = UnsupervisedFacialClassifier(2, 2, 0.01)
train_perf = fc.train(datalist,
epochs,
import wx
from face_sampling_ui import FaceSamplerFrame
from face_file_parser import FaceFileParser
from supervised_facial_classifier import SupervisedFacialClassifier
ffp = FaceFileParser()
ffp.add_dir("./inputData")
# ffp.add_dir("./normData")
datalist = ffp.get_data()
"""
print "Single Hidden Layer"
for i in range(2,20):
fc = FacialClassifier()
fc.alternateTrain(datalist, [(i)], 25)
print "Double Hidden Layer"
for i in range(2,20):
fc = FacialClassifier()
fc.alternateTrain(datalist, (i,i), 25)
print "Triple Hidden Layer"
for i in range(2,20):
fc = FacialClassifier()
fc.alternateTrain(datalist, (i,i,i), 25)
"""
fc = SupervisedFacialClassifier()
# alternateTrain(datalist, hiddenNodes, #epochs, log stats every X epochs)
fc.alternateTrain(datalist, (20, 20), 500, 10)
import wx
from face_sampling_ui import FaceSamplerFrame
from face_file_parser import FaceFileParser
from supervised_facial_classifier import SupervisedFacialClassifier
ffp = FaceFileParser()
ffp.add_dir("./inputData")
#ffp.add_dir("./normData")
datalist = ffp.get_data()
"""
print "Single Hidden Layer"
for i in range(2,20):
fc = FacialClassifier()
fc.alternateTrain(datalist, [(i)], 25)
print "Double Hidden Layer"
for i in range(2,20):
fc = FacialClassifier()
fc.alternateTrain(datalist, (i,i), 25)
print "Triple Hidden Layer"
for i in range(2,20):
fc = FacialClassifier()
fc.alternateTrain(datalist, (i,i,i), 25)
"""
fc = SupervisedFacialClassifier()
# alternateTrain(datalist, hiddenNodes, #epochs, log stats every X epochs)
fc.alternateTrain(datalist, (20,20), 500, 10);
else:
# For all leaf-node directories under the specified path we run trials
for data_dir in os.walk(root_data_dir):
(dirname, subdirs, files) = data_dir
dirs_to_test.append(dirname)
# Write the trial data out to a file
outfile = open(outfile_name, 'w')
outfile.write("DIRNAME, LEARNINGRATE, DECAY, EPOCHS, TRAINING_ERR, TRAINING_ERR_STD, TESTING_ERR, TESTING_ERR_STD\n")
for dirname in dirs_to_test:
for learningrate in learningrates_to_test:
for neighbourdecay in neighbourdecays_to_test:
for epochs in epochs_to_test:
ffp = FaceFileParser(verbose=False)
ffp.add_dir(dirname)
datalist = ffp.get_data()
print "################################"
print " %s, learningrate: %s, decay: %s, epochs: %s" %\
(dirname, learningrate, neighbourdecay, epochs)
trial_avg_testing_err = []
trial_avg_training_err = []
# Run trials
for trial in range(num_trials):
print "Trial #%i" % trial
fc = UnsupervisedFacialClassifier(2,2, 0.01)
train_perf = fc.train(datalist, epochs, test_proportion=0.25, verbose=False)
from face_file_parser import FaceFileParser
from supervised_facial_classifier import SupervisedFacialClassifier
max_hn = 400
min_hn = 50
step_size = 25
epochs = 750
output = 50
ffp = FaceFileParser()
sfc = SupervisedFacialClassifier()
ffp.add_dir('./normData')
d_s = ffp.get_data()
csv_out = open('./vary_topology.csv','w')
csv_out.write('topology,min test error,at epoch,avg test error,avg train error\n')
l1 = min_hn
while l1 <= max_hn:
print 'Topology :' + str(l1) + '\n'
results = sfc.alternateTrain(d_s,[(l1)],epochs,logFreq=output)
min_err = min(results['testing_error'])
at_e = results['epochs'][results['testing_error'].index(min_err)]
avg_tst = results['avg_testing_error']
avg_trn = results['avg_training_error']
csv_out.write(str(l1) + ';' + ',' + str(min_err) + ','
+ str(at_e) + ',' + str(avg_tst) + ',' + str(avg_trn) + '\n')
print 'Minimum error of ' + str(min_err) + ' at epoch ' + str(at_e)
l2 = min_hn
while l2 <= max_hn:
