def process_house_votes_data(file_path):
#print('LOG: process_house_votes_data() START')
#print(file_path)
original_data = FileManager.get_csv_file_data_array(file_path)
#print('LOG: ORIGINAL data')
#print(original_data[0])
#print('LOG: Class moved to final column')
original_col_with_class = 0
columns_moved_data = DataManipulator.move_column_to_end(original_data, original_col_with_class)
#print(columns_moved_data[0])
#print('LOG: group input values into bins')
data_in_bins = _bin_input_attributes(columns_moved_data)
#print(data_in_bins[0])
#print('LOG: turn input features into binary values (0,1)')
num_bins = 3
col_idx = 0
final_data = DataManipulator.expand_attributes_to_binary_values(data_in_bins, col_idx, num_bins)
for col in data_in_bins[0]:
col_idx += num_bins #This is because columns are inserted each itr
#print('length of final_data')
#print(len(final_data[0]))
if col_idx == (len(final_data[0]) - 1): #Skip last col b/c it's the class value
#print('LOG: Stop here - skip the final column which is classification')
break
final_data = DataManipulator.expand_attributes_to_binary_values(final_data, col_idx, num_bins)
#print(final_data[0])
return final_data
def process_house_votes_data(file_path):
#print('LOG: process_house_votes_data() START')
#print(file_path)
original_data = FileManager.get_csv_file_data_array(file_path)
#print('LOG: ORIGINAL data')
#print(original_data[0])
#print('LOG: group input values into bins')
num_bins = 4
#Fix sepal length 4.3 - 7.9
column = 0
min_val = 4.3
max_val = 7.9
data_in_bins = \
_bin_input_attribute(original_data, column, min_val, max_val, num_bins)
#Fix sepal width 2.0 - 4.4
column = 1
min_val = 2.0
max_val = 4.4
data_in_bins = \
_bin_input_attribute(original_data, column, min_val, max_val, num_bins)
#Fix petal length 1.0 - 6.9
column = 2
min_val = 1.0
max_val = 6.9
data_in_bins = \
_bin_input_attribute(original_data, column, min_val, max_val, num_bins)
#Fix petal width 0.1 - 2.5
column = 3
min_val = 0.1
max_val = 2.5
data_in_bins = \
_bin_input_attribute(original_data, column, min_val, max_val, num_bins)
#print()
#print(data_in_bins)
#print()
#print('LOG: turn input features into binary values (0,1)')
col_idx = 0
final_data = DataManipulator.expand_attributes_to_binary_values(
data_in_bins, col_idx, num_bins)
for col in data_in_bins[0]:
col_idx += num_bins #This is because columns are inserted each itr
#print('length of final_data')
#print(len(final_data[0]))
if col_idx == (len(final_data[0]) -
1): #Skip last col b/c it's the class value
#print('LOG: Stop here - skip the final column which is classification')
break
final_data = DataManipulator.expand_attributes_to_binary_values(
final_data, col_idx, num_bins)
#print(final_data[0])
return final_data
def process_house_votes_data(file_path):
#print('LOG: process_house_votes_data() START')
#print(file_path)
original_data = FileManager.get_csv_file_data_array(file_path)
#print('LOG: ORIGINAL data')
#print(original_data[0])
#Eliminate first column of data, convert class from 2,4 -> 0,1 && make bins are 0-9, not 1-10
modified_data = list()
number_of_rows = len(original_data)
for vector_idx in range(0, number_of_rows, 1):
tmplist = original_data[vector_idx][1:]
number_of_cols = len(tmplist)
for col_idx in range(0, number_of_cols, 1):
if col_idx == (number_of_cols - 1): #classificiaton value {2,4}
#Classification value, {2,4} -> {0,1}
if tmplist[col_idx] == 2:
tmplist[col_idx] = 0
elif tmplist[col_idx] == 4:
tmplist[col_idx] = 1
else:
print('ERROR: Expected {2,4} but instead got ',
tmplist[col_idx])
#no modification made
else: #normal bin value
tmplist[col_idx] = tmplist[col_idx] - 1 #bins 1-10 -> 0-9
modified_data.append(tmplist)
#print()
#print(modified_data)
#print()
#print('LOG: turn input features into binary values (0,1)')
num_bins = 10
col_idx = 0
final_data = DataManipulator.expand_attributes_to_binary_values(
modified_data, col_idx, num_bins)
for col in modified_data[0]:
col_idx += num_bins #This is because columns are inserted each itr
#print('length of final_data')
#print(len(final_data[0]))
if col_idx == (len(final_data[0]) -
1): #Skip last col b/c it's the class value
#print('LOG: Stop here - skip the final column which is classification')
break
final_data = DataManipulator.expand_attributes_to_binary_values(
final_data, col_idx, num_bins)
#print(final_data[0])
return final_data
def run_model_with_cross_validation(prune=False):
#GET DATA
#- expect data_0 ... data_4
data_groups = list()
data_groups.append(FileManager.get_csv_file_data_array('data_0'))
data_groups.append(FileManager.get_csv_file_data_array('data_1'))
data_groups.append(FileManager.get_csv_file_data_array('data_2'))
data_groups.append(FileManager.get_csv_file_data_array('data_3'))
data_groups.append(FileManager.get_csv_file_data_array('data_4'))
if prune == True:
validation_data = FileManager.get_csv_file_data_array(
'validation_data')
NUM_GROUPS = len(data_groups)
#For each data_group, train on all others and test on me
culminating_result = 0
culminating_validation_result = 0
finanl_average_result = 0
final_validation_average_result = 0
for test_group_id in range(NUM_GROUPS):
print()
#Form training data as 4/5 data
train_data = list()
for train_group_id in range(len(data_groups)):
if (train_group_id != test_group_id):
#Initialize train_data if necessary
if (len(train_data) == 0):
train_data = data_groups[train_group_id]
else:
train_data = train_data + data_groups[train_group_id]
print('train_data group', str(test_group_id), 'length: ',
len(train_data))
#print(train_data)
test_data = data_groups[test_group_id]
result = 0
validation_result = 0
model = ClassificationTree(train_data)
model.train()
print('tree size:', model.get_size_of_tree())
print('tree: ')
print(model.print_tree())
result = model.test(test_data)
#print('result:', result)
culminating_result = culminating_result + result
print('Accuracy (%):', result)
if prune == True:
model.validate(validation_data)
print('tree size w/ pruning:', model.get_size_of_tree())
print('tree: ')
print(model.print_tree())
validation_result = model.test(test_data)
#print('result:', result)
culminating_validation_result = culminating_validation_result + validation_result
print('Accuracy w/ pruning (%):', validation_result)
print()
final_average_result = culminating_result / NUM_GROUPS
final_validation_average_result = culminating_validation_result / NUM_GROUPS
#print()
#print('final average result:')
#print(final_average_result)
#print()
return (final_average_result, final_validation_average_result)
def main():
#print('LOG: Main program to run tests')
parser = argparse.ArgumentParser(
description='Learn & Verify machine learning algorithms')
parser.add_argument('all_class_filepath',
type=str,
help='full path to input file')
parser.add_argument('class_0_filepath',
type=str,
help='full path to input file')
parser.add_argument('class_1_filepath',
type=str,
help='full path to input file')
parser.add_argument('class_2_filepath',
type=str,
help='full path to input file')
parser.add_argument('fraction',
type=float,
default=0.66,
nargs='?',
help='fraction of data to learn')
parser.add_argument('num_classes',
type=int,
default=3,
nargs='?',
help='number of total classes')
args = parser.parse_args()
#INPUTS
print()
print('INPUTS')
allclass_filepath = args.all_class_filepath
print('overall_model_file_path: ' + allclass_filepath)
class0_filepath = args.class_0_filepath
print('class0_filepath: ' + class0_filepath)
class1_filepath = args.class_1_filepath
print('class1_filepath: ' + class1_filepath)
class2_filepath = args.class_2_filepath
print('class2_filepath: ' + class2_filepath)
fraction_of_data_for_learning = args.fraction
print('fraction of data to train: ', fraction_of_data_for_learning)
number_of_classes = args.num_classes
print('number of classes: ', number_of_classes)
#READ INPUT DATA
allclass_data = FileManager.get_csv_file_data_array(allclass_filepath)
print('number of input vectors: ', len(allclass_data))
class0_data = FileManager.get_csv_file_data_array(class0_filepath)
class1_data = FileManager.get_csv_file_data_array(class1_filepath)
class2_data = FileManager.get_csv_file_data_array(class2_filepath)
#SPLIT INPUT DATA (learning & test sets)
print()
allclass_data_sets = DataManipulator.split_data_in_2_randomly(
allclass_data, fraction_of_data_for_learning)
class0_data_sets = DataManipulator.split_data_in_2_randomly(
class0_data, fraction_of_data_for_learning)
class1_data_sets = DataManipulator.split_data_in_2_randomly(
class1_data, fraction_of_data_for_learning)
class2_data_sets = DataManipulator.split_data_in_2_randomly(
class2_data, fraction_of_data_for_learning)
allclass_learning_data = allclass_data_sets[0]
print('learning data size: ', len(allclass_learning_data))
allclass_test_data = allclass_data_sets[1]
print('test data size: ', len(allclass_test_data))
print()
class0_learning_data = class0_data_sets[0]
class0_test_data = class0_data_sets[1]
class1_learning_data = class1_data_sets[0]
class1_test_data = class1_data_sets[1]
class2_learning_data = class2_data_sets[0]
class2_test_data = class2_data_sets[1]
#LEARN THE MODELS
#Winnow2
winnow2_0 = Winnow2()
winnow2_1 = Winnow2()
winnow2_2 = Winnow2()
winnow2_0_learned_weights = winnow2_0.learn_winnow2_model(
class0_learning_data)
winnow2_1_learned_weights = winnow2_1.learn_winnow2_model(
class1_learning_data)
winnow2_2_learned_weights = winnow2_2.learn_winnow2_model(
class2_learning_data)
print('Winnow2 learned weights for class 0:')
print(winnow2_0_learned_weights)
print()
print('Winnow2 learned weights for class 1:')
print(winnow2_1_learned_weights)
print()
print('Winnow2 learned weights for class 2:')
print(winnow2_2_learned_weights)
print()
#Naive Bayes
naive_bayes = NaiveBayes(number_of_classes)
naive_bayes_learned_percents = naive_bayes.learn_naive_bayes_model(
allclass_learning_data)
print(
'Naive Bayes learned percentages as input[ class[ (prob0, prob1) ] ]')
print(naive_bayes_learned_percents)
print()
#TEST THE MODELS
#Winnow2
print('Testing Winnow2 model')
winnow2_multi_model_test_results = Winnow2.test_multiple_winnow2_models(
allclass_test_data, [winnow2_0, winnow2_1, winnow2_2])
print('classification attempts(', winnow2_multi_model_test_results[0],
'), \
#fails(', winnow2_multi_model_test_results[1], '), \
#success(', winnow2_multi_model_test_results[2], ')')
print()
#Naive Bayes
print('Testing Naive Bayes model')
naive_bayes_test_results = naive_bayes.test_naive_bayes_model(
allclass_test_data)
print('#classification attempts(', naive_bayes_test_results[0], '), \
#fails(', naive_bayes_test_results[1], '), \
#success(', naive_bayes_test_results[2], ')')
print()
def main():
#print('LOG: Main program to run tests')
parser = argparse.ArgumentParser(
description='Learn & Verify machine learning algorithms')
parser.add_argument('file_path', type=str, help='full path to input file')
parser.add_argument('fraction',
type=float,
default=0.66,
nargs='?',
help='fraction of data to learn')
parser.add_argument('num_classes',
type=int,
default=2,
nargs='?',
help='number of total classes')
args = parser.parse_args()
#INPUTS
print()
print('INPUTS')
file_path = args.file_path
print('filepath: ' + file_path)
fraction_of_data_for_learning = args.fraction
print('fraction of data to train: ', fraction_of_data_for_learning)
number_of_classes = args.num_classes
print('number of classes: ', number_of_classes)
#READ INPUT DATA
input_data = FileManager.get_csv_file_data_array(file_path)
print('number of input vectors: ', len(input_data))
#SPLIT INPUT DATA (learning & test sets)
print()
data_sets = DataManipulator.split_data_in_2_randomly(
input_data, fraction_of_data_for_learning)
learning_data = data_sets[0]
print('learning data size: ', len(learning_data))
test_data = data_sets[1]
print('test data size: ', len(test_data))
print()
#LEARN THE MODELS
#Winnow2
winnow2 = Winnow2() # default values for alpha, threshold, & start weight
winnow2_learned_weights = winnow2.learn_winnow2_model(learning_data)
print('Winnow2 learned weights:')
print(winnow2_learned_weights)
print()
#Naive Bayes
naive_bayes = NaiveBayes(number_of_classes)
naive_bayes_learned_percents = naive_bayes.learn_naive_bayes_model(
learning_data)
print(
'Naive Bayes learned percentages as input[ class[ (prob0, prob1) ] ]')
print(naive_bayes_learned_percents)
print()
#TEST THE MODELS
#Winnow2
print('Testing Winnow2 model')
winnow2_test_results = winnow2.test_winnow2_model(test_data)
print('classification attempts(', winnow2_test_results[0], '), \
#fails(', winnow2_test_results[1], '), \
#success(', winnow2_test_results[2], ')')
print()
#Naive Bayes
print('Testing Naive Bayes model')
naive_bayes_test_results = naive_bayes.test_naive_bayes_model(test_data)
print('#classification attempts(', naive_bayes_test_results[0], '), \
#fails(', naive_bayes_test_results[1], '), \
#success(', naive_bayes_test_results[2], ')')
print()
def run_models_with_cross_validation(number_of_layers, nodes_per_layer,
learning_rate, max_epoch, error_thresh):
#GET DATA
#- expect data_0 ... data_4
data_groups = list()
data_type = 'int'
data_groups.append(FileManager.get_csv_file_data_array(
'data_0', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_1', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_2', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_3', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_4', data_type))
model1_culminating_result = 0
model1_final_average_result = 0
NUM_GROUPS = len(data_groups)
#For each data_group, train on all others and test on me
for test_group_id in range(NUM_GROUPS):
print()
#Form training data as 4/5 data
train_data = list()
for train_group_id in range(len(data_groups)):
if (train_group_id != test_group_id):
#Initialize train_data if necessary
if (len(train_data) == 0):
train_data = data_groups[train_group_id]
else:
train_data = train_data + data_groups[train_group_id]
print('train_data group', str(test_group_id), 'length: ',
len(train_data))
#print(train_data)
test_data = data_groups[test_group_id]
test_data = pd.DataFrame(test_data)
train_data = pd.DataFrame(train_data)
print(train_data.head())
model1 = NeuralNetworkFFBP(train_data, number_of_layers,
nodes_per_layer)
model1.train(learning_rate, max_epoch, error_thresh)
print_classifications = False
if (test_group_id == 0
): #Required to print classifications for one fold
print_classifications = True
model1_result = model1.test(
test_data,
print_classifications) # returns (attempts, fails, success)
model1_accuracy = (model1_result[0] / model1_result[1]) * 100
print('Accuracy:', model1_accuracy, '%')
model1_culminating_result = model1_culminating_result + model1_accuracy
model1_final_average_result = model1_culminating_result / NUM_GROUPS
#print()
#print('final average result:')
#print(final_average_result)
#print()
return (model1_final_average_result)
def run_models_with_cross_validation(num_classes=2, learning_rate=0.5):
#GET DATA
#- expect data_0 ... data_4
data_groups = list()
data_type = 'int'
data_groups.append(FileManager.get_csv_file_data_array(
'data_0', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_1', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_2', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_3', data_type))
data_groups.append(FileManager.get_csv_file_data_array(
'data_4', data_type))
NUM_GROUPS = len(data_groups)
#For each data_group, train on all others and test on me
model1_culminating_result = 0
model2_culminating_result = 0
model1_final_average_result = 0
model2_final_average_result = 0
for test_group_id in range(NUM_GROUPS):
print()
#Form training data as 4/5 data
train_data = list()
for train_group_id in range(len(data_groups)):
if (train_group_id != test_group_id):
#Initialize train_data if necessary
if (len(train_data) == 0):
train_data = data_groups[train_group_id]
else:
train_data = train_data + data_groups[train_group_id]
print('train_data group', str(test_group_id), 'length: ',
len(train_data))
#print(train_data)
test_data = data_groups[test_group_id]
model1_result = 0
model2_result = 0
model1 = NaiveBayes(num_classes)
model2 = LogisticRegression(pd.DataFrame(train_data))
model1.train(train_data)
model2.train(pd.DataFrame(train_data), learning_rate)
print_classifications = False
if (test_group_id == 0
): #Required to print classifications for one fold
print_classifications = True
model1_result = model1.test(
test_data,
print_classifications) # returns (attempts, fails, success)
#print('result:', result)
model1_accuracy = (model1_result[2] / model1_result[0]) * 100
print('Naive Bayes Accuracy (%):', model1_accuracy)
model2_result = model2.test(
pd.DataFrame(test_data),
print_classifications) # returns (% accuracy)
print('Logistic Regression Accuracy (%):', model2_result)
model1_culminating_result = model1_culminating_result + model1_accuracy
model2_culminating_result = model2_culminating_result + model2_result
model1_final_average_result = model1_culminating_result / NUM_GROUPS
model2_final_average_result = model2_culminating_result / NUM_GROUPS
#print()
#print('final average result:')
#print(final_average_result)
#print()
return (model1_final_average_result, model2_final_average_result)
