def test_two_sentence_language(self):
test_sequences = ['a b c d.', 'e f g.']
N = 2
sampler = utilities.SentenceSamplerUtility(test_sequences, N)
sample = sampler.get_sample()
print(sample)
assert sample in test_sequences
def build_samples(class_value, y_value, n, training_file, generate_html):
'''
:param class_attr: Column name in train csv that specifies class
:param file_name_attr: Column name in train csv that specifies file name
:param class_value: Name of class (In this case unsponsored and sponsored)
:param y_value: The y value of the class (In this case 0 - unsponsored and 1 - sponsored)
:param n: Number of sample files
:param training_file: Name of the train csv
:param generate_html: Boolean value to specify whether we want to generate the HTML file for the sampled html text files
:return: A list of file names that get sampled. This is important because not all files sampled are available in the data.
'''
class_files = sampler.get_rows_of_class(training_file, y_value)
sampled_files = sampler.get_sample(class_files, n)
return sampler.collect_sample_data(sampled_files, class_value, generate_html=generate_html)
def build_samples(class_value, y_value, n, training_file, generate_html):
'''
:param class_attr: Column name in train csv that specifies class
:param file_name_attr: Column name in train csv that specifies file name
:param class_value: Name of class (In this case unsponsored and sponsored)
:param y_value: The y value of the class (In this case 0 - unsponsored and 1 - sponsored)
:param n: Number of sample files
:param training_file: Name of the train csv
:param generate_html: Boolean value to specify whether we want to generate the HTML file for the sampled html text files
:return: A list of file names that get sampled. This is important because not all files sampled are available in the data.
'''
class_files = sampler.get_rows_of_class(training_file, y_value)
sampled_files = sampler.get_sample(class_files, n)
return sampler.collect_sample_data(sampled_files,
class_value,
generate_html=generate_html)
def train(train_file, parameters):
model = Model(parameters)
batch = parameters["batch"]
niter = parameters["niter"]
sample = sampler.get_sample(parameters)
logger = logging.getLogger(Logger.project_name)
##initilization the sampler
train_reader = SvmReader(train_file, batch)
has_next = True
while has_next:
x,y,has_next = train_reader.read()
sample.update(y)
##weight updates
for iter1 in xrange(niter):
train_reader = SvmReader(train_file, batch)
has_next = True
while has_next:
x, y, has_next = train_reader.read()
idx = sample.sample(y)
model.update(x, y, idx)
logger.info("The %d-th iteration completes"%(iter1+1));
train_reader.close()
##tuning threshold
train_reader = SvmReader(train_file, batch)
x, y, has_next = train_reader.read()
while has_next:
p = model.ff(x)
model.thrsel.update(p, y)
x, y, has_next = train_reader.read()
return model
def test_utilities_sampler_construction(self):
test_sequences = ['a b c d e f g.']
N = 2
sampler = utilities.SentenceSamplerUtility(test_sequences, N)
assert sampler.get_sample() == 'a b c d e f g.'
def train_mem(train_file, parameters):
model = Model(parameters)
batch = parameters["batch"]
niter = parameters["niter"]
sample = sampler.get_sample(parameters)
logger = logging.getLogger(Logger.project_name)
logger.info("Model initialization done")
train_reader = SvmReader(train_file)
x, y = train_reader.full_read()
num, _ = y.shape
#if None == sample: idx_y = sp.csr_matrix(np.ones(y.shape))
#else: idx_y = sp.csr_matrix(sample.sample(y))
logger.info("Training data loading done")
sample.update(y)
logger.info("Sampling initialization done")
start_time = time.time()
for iter1 in xrange(niter):
start = 0
end = batch
while start < num:
#logger.info("start = %d, end = %d\n"%(start, end))
if end > num: end = num
# import cProfile, pstats, StringIO
# pr = cProfile.Profile()
# pr.enable()
batch_x = x[start:end, :]
batch_y = y[start:end, :]
batch_i = sample.sample(batch_y)
model.update(batch_x, batch_y, batch_i)
start += batch;
end += batch;
# pr.disable()
# s = StringIO.StringIO()
# sortby = 'cumulative'
# ps = pstats.Stats(pr, stream = s).sort_stats(sortby)
# ps.print_stats()
# print "update",s.getvalue()
logger.info("The %d-th iteration completes"%(iter1+1));
#####tuning the threshold
total = 0
start = 0
end = batch
while start < num and total < 1000:
if end > num: end = num
batch_x = x[start:end,:]
batch_y = y[start:end,:]
batch_p = model.ff(batch_x)
model.thrsel.update(batch_p, batch_y)
start += batch
end += batch
total += 1
logger.info("The threshold tuning completes")
end_time = time.time()
logger.info("The training time is %f"%(end_time-start_time))
return model
