def create_label_after_das(da_output_seq, algo, rank, window):
if algo == "avg_disp":
d_x = create_av_disp(
sequence = da_output_seq,
rank = rank,
window_size = window
)
elif algo == "avg":
d_x = create_av(
sequence = da_output_seq,
rank = rank,
window_size = window
)
elif algo == "filter+avg":
d_x = create_av(
sequence = filter_data(da_output_seq),
rank = rank,
window_size = window
)
elif algo == "filter+avg_disp":
d_x = create_av_disp(
sequence = filter_data(da_output_seq),
rank = rank,
window_size = window
)
else:
d_x = [numpy.mean(create_int_labels(
sequence = da_output_seq,
rank = rank
))]
return d_x[0]
def read_all(self):
# sequence_matrix = array[size of 1st doc][ data.z, data.gt]
sequence_matrix = self.get_sequence()
# d_x1 = array[size of 1st doc][z]
d_x1 = filter_data(
sequence = sequence_matrix[:, self.n_in-1]
)
# d_y1 = array[size of 1st doc][labels]
d_y1 = sequence_matrix[:, self.n_in]
# data_x_ar = union for z-coordinate in all files
data_x = d_x1
# data_y_ar = union for labels in all files
data_y = d_y1
for t in range(len(self.seqs) - 1):
# sequence_matrix = array[size of t-th doc][data.z, data.gt]
sequence_matrix = self.get_sequence()
# d_x = array[size of t-th doc][z]
d_x = filter_data(
sequence = sequence_matrix[:, self.n_in-1]
)
# d_y = array[size of t-th doc][labels]
d_y = sequence_matrix[:, self.n_in]
# concatenate data in current file with data in prev files in one array
data_x = numpy.concatenate((data_x, d_x))
data_y = numpy.concatenate((data_y, d_y))
gc.collect()
set_x = theano.shared(numpy.asarray(data_x,
dtype=theano.config.floatX),
borrow=True)
set_y = T.cast(theano.shared(numpy.asarray(data_y,
dtype=theano.config.floatX),
borrow=True), 'int32')
return (set_x, set_y)
def read_next_doc(self, algo, rank=1, window=1, divide = False):
# sequence_matrix = array[size of doc][data.z, data.gt]
sequence_matrix = self.get_sequence()
# d_x = array[size of doc][z]
if algo == "filter":
d_x = filter_data(
sequence = sequence_matrix[:, self.n_in-1]
)
elif algo == "normalize_1_1":
d_x = normalize_sequence_1_1(
sequence = sequence_matrix[:, self.n_in-1]
)
elif algo == "normalize_0_1":
d_x = normalize_sequence_0_1(
sequence = sequence_matrix[:, self.n_in-1]
)
elif algo == "int_labels":
d_x = create_int_labels(
sequence = sequence_matrix[:, self.n_in-1],
rank = rank
)
elif algo == "avg_disp":
d_x = create_av_disp(
sequence = sequence_matrix[:, self.n_in-1],
rank = rank,
window_size = window
)
elif algo == "avg":
d_x = create_av(
sequence = sequence_matrix[:, self.n_in-1],
rank = rank,
window_size = window
)
elif algo == "filter+avg":
d_x = create_av(
sequence = filter_data(sequence_matrix[:, self.n_in-1]),
rank = rank,
window_size = window
)
elif algo == "filter+avg_disp":
d_x = create_av_disp(
sequence = filter_data(sequence_matrix[:, self.n_in-1]),
rank = rank,
window_size = window
)
else:
d_x = sequence_matrix[:, self.n_in-1]
# d_y = array[size of doc][labels]
d_y = sequence_matrix[:, self.n_in]
d_y = d_y[window/2: len(d_y) + window/2 -window +1]
gc.collect()
if not divide:
set_x = theano.shared(numpy.asarray(d_x),
borrow=True)
set_y = T.cast(theano.shared(numpy.asarray(d_y,
dtype=theano.config.floatX),
borrow=True), 'int32')
return (set_x, set_y)
data = zip(d_x, d_y) #pairs (modified z-coord, label)
visible_seqs = []
for label in xrange(7):
d_x_for_label=[]
for row in data:
if row[-1] == label:
d_x_for_label.append(row[0])
set_x = theano.shared(numpy.asarray(d_x_for_label,
dtype=theano.config.floatX),
borrow=True)
visible_seqs.append(set_x)
return visible_seqs
