def writeArffForInclusiveSubset(filename, data, attributes, subset):
num_attrs = len(attributes)
assert(len(subset) <= num_attrs)
assert(len(subset) >= 2)
for d in data:
assert(len(d) == num_attrs)
attrs_subset = [attributes[i] for i in range(num_attrs) if i in subset]
data_subset = [[d[i] for i in range(num_attrs) if i in subset] for d in data]
if False:
print 'subset = ', len(subset), subset
print 'num_attrs =', num_attrs
print 'attributes =', len(attributes), [a['name'] for a in attributes]
print 'attrs_subset =', len(attrs_subset), [a['name'] for a in attrs_subset]
assert(len(attrs_subset) >= 2)
arff.writeArff(filename, None, 'find_best_attr', attrs_subset, data_subset)
def makeTrainingTestSplit(base_data, split_vector, prefix):
""" Split <base_data> into training and test data sets. Rows with indexes in
<split_vector> go into training file and remaining go into test file.
Writes training and test .arff files and returns their names.
File names are prefixed with <prefix> """
assert len(base_data) == len(split_vector)
num_instances = len(base_data)
training_file_name = prefix + training_file_base
test_file_name = prefix + test_file_base
training_data = []
test_data = []
for i, x in enumerate(base_data):
if split_vector[i]:
test_data.append(x)
else:
training_data.append(x)
arff.writeArff(training_file_name, base_comments, base_relation, base_attrs, training_data)
arff.writeArff(test_file_name, base_comments, base_relation, base_attrs, test_data)
return (training_file_name, test_file_name)
print 'Usage: jython get_attribute_subset.py <base-arff-file> <attrs-arff-file>'
sys.exit()
base_filename = sys.argv[1]
attrs_filename = sys.argv[2]
out_filename = os.path.splitext(base_filename)[0] + '.attr_subset' + os.path.splitext(base_filename)[1]
print base_filename
print attrs_filename
print out_filename
relation, comments, attributes, data = arff.readArff(base_filename)
_, _, attributes_subset, _ = arff.readArff(attrs_filename)
attribute_index_map = {}
for i,a in enumerate(attributes):
attribute_index_map[a['name']] = i
names_subset = [a['name'] for a in attributes_subset]
indexes_subset = []
for name in attribute_index_map.keys():
if name in names_subset:
indexes_subset.append(attribute_index_map[name])
out_attributes = [attributes[i] for i in indexes_subset]
out_data = [[d[i] for i in indexes_subset] for d in data]
arff.writeArff(out_filename, comments, relation, out_attributes, out_data)
attrs_to_combine = [] remaining_attrs = [] for val in attributes[class_index]['vals']: if val in classes_to_combine: attrs_to_combine.append(val) else: remaining_attrs.append(val) combine_attributes = copy.deepcopy(attributes) combine_attributes[class_index]['vals'] = [group_name] + remaining_attrs separate_attributes = copy.deepcopy(attributes) separate_attributes[class_index]['vals'] = attrs_to_combine data_to_combine = [] separate_data = [] remaining_data = [] for d in data: if d[class_index] in classes_to_combine: separate_data.append(d) d2 = copy.deepcopy(d) d2[class_index] = group_name data_to_combine.append(d2) else: remaining_data.append(d) combine_data = data_to_combine + remaining_data arff.writeArff(combine_filename, comments, relation, combine_attributes, combine_data) arff.writeArff(separate_filename, comments, relation, separate_attributes, separate_data)
