def filter_and_store(files, dr, f_name):
ch = my_chain.Chain(
my_chain.ChainElement(my_tools.ReadSignal, [{
'files': [files]
}]),
#my_chain.ChainElement(my_tools.Filter,
# [{'wp':[0.5, 30], 'ws':[0.1, 45.0], 'gpass': 5.0,
# 'gstop': 40.0, 'ftype':'cheby2', 'unit':'hz', 'use_filtfilt':False}]),
#my_chain.ChainElement(my_tools.Filter,
# [{'wp':0.8, 'ws':0.2, 'gpass': 2.0,
# 'gstop': 60.0, 'ftype':'ellip', 'unit':'hz', 'use_filtfilt':True}]),
my_chain.ChainElement(my_tools.Filter, [{
'wp': 20.0,
'ws': 30.0,
'gpass': 5.0,
'gstop': 60.0,
'ftype': 'cheby2',
'unit': 'hz',
'use_filtfilt': True
}]),
my_chain.ChainElement(my_tools.SaveToFile, [{
'dir_path': dr,
'file_name': f_name
}]))
cs, res = ch.process(None, False, False)
def run():
dr2 = '/media/windows/wiedza/bci/EKSPERYMENTY_DANE/p300_10_12_2010/squares/'
f2_name = 'p300_128hz_laptop_training_6x6_square_CATDOGFISHWATERBOWL_longer_8trials2'
f2 = {
'info': os.path.join(dr2, f2_name+'_10HZ.obci.xml'),
'data': os.path.join(dr2, f2_name+'_10HZ.obci.bin'),
'tags':os.path.join(dr2, f2_name+'.obci.arts_free.svarog.tags')
}
"""dr2 = '/media/windows/wiedza/bci/EKSPERYMENTY_DANE/p300_10_12_2010/numbered_squares/'
f2_name = 'p300_128hz_laptop_training_6x6_squareNUMBERS_CATDOGFISHWATERBOWL_longer_8trials'
f2 = {
'info': os.path.join(dr2, f2_name+'.obci.filtered.xml'),
'data': os.path.join(dr2, f2_name+'_10HZ.obci.bin'),
'tags':os.path.join(dr2, f2_name+'.obci.arts_free.svarog.tags')
}"""
mgr = read_manager.ReadManager(f2['info'], f2['data'], f2['tags'])
train_data_ch, train_labels = prepare.get_train_set_from_mgr(mgr, num_per_avg=10, start_samples_to_norm=0, downsample_level=5)
class MY_SVM(object):
def __init__(self, C):
self.s = svm.SVM(C=C)
def process(self, data):
return self.s.stratifiedCV(data, 5).getBalancedSuccessRate()
def __repr__(self):
return "SVM: C - "+str(self.s.C)
class MY_STD(object):
def process(self, data):
#p = prepare.Standardizer()
#p.train(data)
data.normalize(2)
return data
ch = my_chain.Chain(
my_chain.ChainElement(MY_STD, {}),
my_chain.ChainElement(MY_SVM,
{'C':[0.01, 0.1, 0.5, 1, 10]})
)
l = Labels([str(i) for i in train_labels])
for i, train_data in enumerate(train_data_ch):
train_data = train_data_ch[11]
data = VectorDataSet(train_data, L=l)
cs, res = ch.process(data)
print("CSS: "+str(cs))
print("RES: "+str(res))
break
def run(files, folds=3, avg_size=10):
ch = my_chain.Chain(
my_chain.ChainElement(my_tools.ReadSignal, {'files': [files]}),
#my_chain.ChainElement(my_tools.ExcludeChannels,
# {'channels':[#['SAMPLE_NUMBER', 'F8'],
# ['SAMPLE_NUMBER', 'F8', 'M1', 'M2']
# ]
# }),
#my_chain.ChainElement(my_tools.Plot,
# [{'channel':'Pz'}]),
my_chain.ChainElement(
my_tools.Montage,
[
{
'montage_type': 'no_montage'
},
#{'montage_type': 'common_spatial_average'},
{
'montage_type': 'ears',
'l_ear_channel': 'M1',
'r_ear_channel': 'M2'
}
]),
my_chain.ChainElement(
my_tools.LeaveChannels,
{'channels': [
['Cz'],
['C3'],
['C4'],
['Fz'],
['Pz'],
]}),
my_chain.ChainElement(my_tools.Filter, [{
'wp': 0.3,
'ws': 0.1,
'gpass': 3.0,
'gstop': 30.0,
'ftype': 'cheby2',
'unit': 'hz'
}]),
my_chain.ChainElement(my_tools.Filter, [{
'wp': 15.0,
'ws': 25.0,
'gpass': 1.0,
'gstop': 60.0,
'ftype': 'ellip',
'unit': 'hz'
}]),
my_chain.ChainElement(
my_tools.Segment, {
'classes': [('target', 'non-target')],
'start_offset': [-0.2, -0.1, 0.0, 0.1, 0.15],
'duration': [0.4, 0.5, 0.6]
}),
my_chain.ChainElement(
my_tools.Average, {
'bin_selectors': [[
lambda mgr: mgr['name'] == 'target',
lambda mgr: mgr['name'] == 'non-target'
]],
'bin_names': [['target', 'non-target']],
'size': [avg_size],
'baseline': [0.0, 0.1, 0.2],
'strategy': ['random']
}),
my_chain.ChainElement(
my_tools.Downsample, # odwrotnie?
{'factor': [2, 3, 4, 5, 7, 9]}),
my_chain.ChainElement(my_tools.Normalize, {'norm': [2]}),
my_chain.ChainElement(my_tools.PrepareTrainSet, {}),
my_chain.ChainElement(
my_tools.SVM, {
'C': [1.0, 3.0, 5.0, 10.0],
'Cmode': ['classProb'],
'folds': [folds],
'kernel': [ker.Linear()]
}))
cs, res = ch.process(None, True, True)
#now make svm better
ret = files
for elem in cs[:-1]:
ret = elem.process(ret)
new_ch = my_chain.Chain(
my_chain.ChainElement(
my_tools.SVM, {
'C': [
0.01, 0.05, 0.1, 0.2, 0.5, 1.0, 2.0, 3.0, 5.0, 7.0, 10.0,
12.0, 15.0, 20.0
],
'Cmode': ['classProb'],
'folds': [folds],
'kernel': [
ker.Gaussian(1.0),
ker.Gaussian(3.0),
ker.Gaussian(5.0),
ker.Gaussian(7.0),
ker.Gaussian(10.0),
ker.Gaussian(2.0),
ker.Gaussian(4.0),
ker.Gaussian(5.0),
ker.Gaussian(8.0),
ker.Gaussian(11.0),
ker.Gaussian(12.0),
ker.Gaussian(15.0),
ker.Gaussian(18.0),
ker.Gaussian(20.0),
ker.Gaussian(25.0),
ker.Polynomial(3),
ker.Polynomial(2),
ker.Linear()
]
}))
new_cs, new_res = new_ch.process(ret, True, True)
cs[-1] = new_cs[-1]
print("#############################################")
print("CANDIDATES LEN: " + str(len(ch.candidates)))
print("#############################################")
ch.print_errors()
ch.print_candidate(cs)
print("#############################################")
print("RESULT1: " + str(res))
print("RESULT2: " + str(new_res))
return cs, new_res
import sys
import p300_sample_data
if __name__ == '__main__':
person = sys.argv[1]
mode = sys.argv[2]
folds = int(sys.argv[3])
avg_size = int(sys.argv[4])
files = p300_sample_data.get_files(person, mode)
cs, res = run(files, folds, avg_size)
cs[-1].ret = 'result'
r = files
for elem in cs:
r = elem.process(r)
ch = my_chain.Chain()
ch.print_candidate(cs)
print("Training/Testing result: " + str(res) + "/" +
(str(r.getBalancedSuccessRate())))
for i in range(cs[-1].folds):
print("##########")
print(r[i])
print("*******")
print(r)
r.plotROC()
def run():
dr2 = '/media/windows/wiedza/bci/EKSPERYMENTY_DANE/p300_10_12_2010/squares/'
f2_name = 'p300_128hz_laptop_training_6x6_square_CATDOGFISHWATERBOWL_longer_8trials2'
f2 = {
'info': os.path.join(dr2, f2_name + '_10HZ.obci.xml'),
'data': os.path.join(dr2, f2_name + '_10HZ.obci.bin'),
'tags': os.path.join(dr2, f2_name + '.obci.arts_free.svarog.tags')
}
"""dr2 = '/media/windows/wiedza/bci/EKSPERYMENTY_DANE/p300_10_12_2010/numbered_squares/'
f2_name = 'p300_128hz_laptop_training_6x6_squareNUMBERS_CATDOGFISHWATERBOWL_longer_8trials'
f2 = {
'info': os.path.join(dr2, f2_name+'.obci.filtered.xml'),
'data': os.path.join(dr2, f2_name+'_10HZ.obci.bin'),
'tags':os.path.join(dr2, f2_name+'.obci.arts_free.svarog.tags')
}"""
#train_data_ch, train_labels = prepare.get_train_set(f2, num_per_avg=15, start_samples_to_norm=0, downsample_level=5)
class MY_SVM(object):
def __init__(self, C, Cmode, kernel):
self.s = svm.SVM(C=C, Cmode=Cmode, arg=kernel)
def process(self, data):
return self.s.stratifiedCV(data[0], 5).getBalancedSuccessRate()
def __repr__(self):
return str({
"CLASS": self.__class__.__name__,
"Cmode": self.s.Cmode,
"kernel": self.s.kernel.__class__.__name__,
"s.C": str(self.s.C)
})
class MY_PREPARE(object):
def __init__(self, num_per_avg, start_samples_to_norm,
downsample_level, start_sec_offset, duration):
self.num_per_avg = num_per_avg
self.start_samples_to_norm = start_samples_to_norm
self.downsample_level = downsample_level
self.start_sec_offset = start_sec_offset
self.duration = duration
def process(self, mgrs):
train_data_ch, train_labels = prepare.get_train_set_from_mgr(
mgrs[0], self.num_per_avg, self.start_samples_to_norm,
self.downsample_level, self.start_sec_offset, self.duration)
l = Labels([str(i) for i in train_labels])
data = VectorDataSet(train_data_ch[0], L=l)
data.normalize(2)
return [data]
def __repr__(self):
return str({
"CLASS": self.__class__.__name__,
"num_per_avg": self.num_per_avg,
"start_samples_to_norm": self.start_samples_to_norm,
"duration": self.duration,
"start_sec_offset": self.start_sec_offset,
"downsample_level": self.downsample_level
})
ch = my_chain.Chain(
my_chain.ChainElement(my_tools.ReadSignal, {'files': [f2]}),
my_chain.ChainElement(
my_tools.ExcludeChannels, {
'channels': [['SAMPLE_NUMBER', 'F8'],
['SAMPLE_NUMBER', 'F8', 'M1', 'M2']]
}),
my_chain.ChainElement(my_tools.Montage,
[{
'montage_type': 'no_montage'
}, {
'montage_type': 'common_spatial_average'
}, {
'montage_type': 'ears',
'l_ear_channel': 'M1',
'r_ear_channel': 'M2'
}]),
my_chain.ChainElement(
my_tools.LeaveChannels,
{
'channels': [
['Cz'],
#['C3'],
['Fz'],
['Pz'],
]
}),
my_chain.ChainElement(
MY_PREPARE, {
'num_per_avg': [10],
'start_samples_to_norm': [0, 25, 50],
'downsample_level': [3, 5, 7, 10],
'start_sec_offset': [-0.2, 0.0, 0.1],
'duration': [0.4, 0.6]
}),
#my_chain.ChainElement(my_tools.Normalize,
# {'norm': [2]}),
my_chain.ChainElement(
MY_SVM, {
'C': [0.5, 1, 3, 5, 7, 10],
'Cmode': ['classProb'],
'kernel': [ker.Linear(), ker.Polynomial()]
}))
cs, res = ch.process(None, True)
print("#############################################")
print("CANDIDATES LEN: " + str(len(ch.candidates)))
print("#############################################")
ch.print_errors()
ch.print_candidate(cs)
print("#############################################")
print("RESULT: " + str(res))