def getAccuracyForInclusiveSubset(output_dir, algo_key, data, attributes, subset):
""" Return accuracy for algorithm with name <algo_key> on <data> and
attributes <attributes> for the attribute subset <subset>
"""
assert(len(attributes) >= 2)
training_filename = makeFileName(output_dir,'find-best-attr', algo_key, None, 'arff')
writeArffForInclusiveSubset(training_filename, data, attributes, subset)
result = WC.getAccuracyAlgoKey(algo_key, class_index, training_filename)
misc.rm(training_filename)
return result
def getAccuracyForSplit(base_data, split_vector):
""" 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.
Run prediction and return accuracy
"""
training_filename, test_filename = makeTrainingTestSplit(base_data, split_vector, "temp")
accuracy = getAccuracy(training_filename, test_filename)
misc.rm(training_filename)
misc.rm(test_filename)
return accuracy
def getSubsetResultDictList(algo_key, data, attributes, exclusive_subset_list):
""" Returns a list of dicts that shows results of running <algo_key> classfier on
<data> for (complement of x) <attributes) for x in <exclusive_subset_list>
"""
assert(None not in exclusive_subset_list)
number = len(exclusive_subset_list)
inclusive_subset_list = [getInclusiveSubset(attributes, subset, is_inclusive) for subset in exclusive_subset_list]
training_filename_list = [makeFileName(output_dir, 'find-best-attr%02d'% i, algo_key, None, 'arff') for i in range(number)]
for i in range(number):
writeArffForInclusiveSubset(training_filename_list[i], data, attributes, inclusive_subset_list[i])
score_eval_list = WC.getAccuracyAlgoKeyList(algo_key, class_index, training_filename_list)
#print 'score_eval_list', score_eval_list
for training_filename in training_filename_list:
misc.rm(training_filename)
result_list = [{'subset':exclusive_subset_list[i], 'score':score_eval_list[i][0], 'eval': score_eval_list[i][1]} for i in range(number)]
if verbose or show_scores:
for result in result_list:
print 'getSubsetResultDict =>', result['score'], result['subset']
return result_list
