def test_findSplit_easy(self):
data = load_testdata('fake_dataset_small.txt')
split = DecisionTree.findSplit(None, data)
self.assertEqual(split, (2, -15), "Should be (2,-15)")
left, right = DecisionTree.partitionData(self, data, split)
self.assertEqual(DecisionTree.findSplit(None, left), (None, None),
"Should be (None,None)")
self.assertEqual(DecisionTree.findSplit(None, right), (4, -20),
"Should be (4,-20)")
def load_tree(file, manager):
global CURRENT_TREE
f = open(file, 'r')
CURRENT_TREE = file
if not f.read():
f.close()
tree = DecisionTree(manager, file, True)
# ready for training
else:
f.close()
tree = DecisionTree(manager, file)
# tree.print_tree()
return tree
def run(params, manager):
global TREE
command = params[0]
options = params[1]
if command == 'train':
# train
if len(params) != 3:
bad_input()
else:
dimension = params[2]
file = options
if os.path.isfile(file):
manager.train(file, TREE, dimension)
else:
print("file %s does not exists" % file)
TREE.save_tree(CURRENT_TREE)
elif command == 'play':
# start game
numbered = False
if len(params) > 2:
if params[2] == '-n':
numbered = True
dimension = options
# TREE.print_tree()
Game(TREE, manager, dimension, numbered)
TREE.save_tree(CURRENT_TREE)
TREE = DecisionTree(manager, CURRENT_TREE)
elif command == 'tree':
# load or view
if options == '-l':
# load tree
file = params[2]
if os.path.isfile(file):
new_tree = load_tree(file, manager)
return new_tree
else:
print("file %s does not exist." % file)
elif options == '-p':
TREE.print_tree()
else:
print("tree command unrecognized")
print("tree -l - load a new decision tree")
print("tree -p - print the current decision tree\n")
pass
else:
bad_input()
return TREE
def test_predict(self):
df = pd.DataFrame(
{
'YearMade': [1984, 1986, 2000, 1984, 1986, 2000, 1984, 1986],
'Price': [2000, 2000, 4000, 2000, 2000, 4000, 2000, 2000]
},
index=[0, 1, 2, 3, 4, 5, 6, 7])
x = df.iloc[:, 0:1]
y = np.array(df['Price'])
tree = DecisionTree(x, y)
x_test = pd.DataFrame({'YearMade': [1985, 1985, 2002]},
index=[0, 1, 2])
test_predictions = tree.predict(x_test)
expected_predictions = [2000, 2000, 4000]
self.assertEqual(
(np.array(test_predictions) == np.array(expected_predictions)
).all(), True)
def test_partition_data(self):
data = load_testdata('fake_dataset_small.txt')
split = (2, -15)
left, right = DecisionTree.partitionData(self, data, split)
np.testing.assert_array_equal(left, data[data[:, split[0]] < split[1]],
"Split arrays do not match")
np.testing.assert_array_equal(right,
data[data[:, split[0]] >= split[1]],
"Split arrays do not match")
def test_setupTree(self):
decisiontree = DecisionTree(self.data)
self.assertEqual(
decisiontree.xm, self.x_train.shape[1],
"Should be " + str(self.x_train.shape[1]) + " attributes")
self.assertEqual(
decisiontree.xn, self.x_train.shape[0],
"Should be " + str(self.x_train.shape[0]) + " datapoints")
np.testing.assert_array_equal(decisiontree.data, self.data,
"x_train data does not match")
delimiter = ';'
metadata = ""
seed = None
for i in range(len(argv)):
if (argv[i] in ['--file', '-f']):
filename = argv[i + 1]
elif (argv[i] in ['--delimiter', '-d']):
delimiter = argv[i + 1]
elif (argv[i] in ['--meta', '-m']):
metadata = argv[i + 1]
elif (argv[i] in ['--seed', '-s']):
seed = int(argv[i + 1])
elif (argv[i] in ['--help', '-h']):
print_usage(argv[0])
exit(0)
if seed is not None:
random.seed(int(seed))
# load the dataset to memory
db = Dataset(filename, delimiter=delimiter, metadata=metadata)
# generates a decision tree from the dataset
tree = DecisionTree(db.data,
db.attributes,
db.target_attribute,
db.numeric,
single_tree_print=True)
# print the resultant tree on the terminal
print(tree)
def cross_val(data, folds):
start = datetime.datetime.now()
splits = crossValidation_split(data, folds)
stats = []
stats_pruned = []
#Creats a tree and evaluates for every fold. Results collected in list "stats"
for i in range(folds):
#Take one testset
test = data[splits[i], :]
results_lowlevel_unpruned = []
results_lowlevel_pruned = []
#Take only non-test folds for train and val
remaining_folds = np.delete(np.arange(folds), i)
for j in remaining_folds:
mask_train = np.ones(data.shape[0], dtype=bool)
mask_train[splits[i]] = False
mask_train[splits[j]] = False
train = data[mask_train, :]
val = data[splits[j]]
decisiontree = DecisionTree()
decisiontree.buildTree(train)
results_lowlevel_unpruned.append(decisiontree.evaluate(test))
decisiontree.prune(train, val)
results_lowlevel_pruned.append(decisiontree.evaluate(test))
print("Tree Complete! Test Set: {} Validation Set: {}".format(
i, j))
stats.append(average_statistics(results_lowlevel_unpruned))
stats_pruned.append(average_statistics(results_lowlevel_pruned))
result = average_statistics(stats)
result_pruned = average_statistics(stats_pruned)
end = datetime.datetime.today()
print("")
print("#############################")
print(
"{}-fold Crossvalidation complete. Average scores unpruned tree over all folds:"
.format(folds))
print("Average Confusion Matrix:")
print(result['confusionmatrix'])
print("Average Precision: {:.2%}".format(np.mean(result['precision'])))
print("Average Precision per Class:")
print(result['precision'])
print("Average Recall: {:.2%}".format(np.mean(result['recall'])))
print("Average Recall per Class:")
print(result['recall'])
print("Average F1: {:.2%}".format(np.mean(result['F1score'])))
print("Average F1 score per Class:")
print(result['F1score'])
print("Average Classification Rate: {:.2%}".format(result['posClassRate']))
print("#############################")
print("Average scores pruned tree over all folds:")
print("Average Confusion Matrix:")
print(result_pruned['confusionmatrix'])
print("Average Precision: {:.2%}".format(
np.mean(result_pruned['precision'])))
print("Average Precision per Class:")
print(result_pruned['precision'])
print("Average Recall: {:.2%}".format(np.mean(result_pruned['recall'])))
print("Average Recall per Class:")
print(result_pruned['recall'])
print("Average F1 score: {:.2%}".format(np.mean(result_pruned['F1score'])))
print("Average F1 score per Class:")
print(result_pruned['F1score'])
print("Average Classification Rate: {:.2%}".format(
result_pruned['posClassRate']))
print("Runtime: {}".format(end - start))
print("#############################")
print("")
return result, result_pruned
def train(self, training_set):
for train_set in training_set:
self.Forest.append(
DecisionTree(train_set, self.db.attributes,
self.db.target_attribute, self.db.numeric))
# Random Forest / Naiive Bayes
from src.loader import DataFactory
from src.DecisionTree import DecisionTree
from src.RandomForest import RandomForest
import numpy as np
import matplotlib.pyplot as plt
df = DataFactory()
animal_df = df.get_dataframe()
dt = DecisionTree(animal_df)
dt.run_simulation('class_type')
rf = RandomForest(animal_df)
rf.run_simulation('class_type')
def test_score_split(self):
tree = DecisionTree(self.x, self.y)
score, _, _ = tree.score_split(1, 0)
self.assertEqual(score, 1000)
def test_find_best_split(self):
tree = DecisionTree(self.x, self.y)
self.assertEqual(tree.split_row, 1985)
def test_is_leaf(self):
tree = DecisionTree(self.x, self.y)
self.assertEqual(tree.left.is_leaf(), True)
cnt_cutoff = 3
gain_cutoff = 0.1
train_ac = []
test_ac = []
for i in range(k):
idx = np.arange(l)
cv_set = indexes[int(i * l / k):int((i + 1) * l / k)]
train_data = data[np.logical_not(np.isin(idx, cv_set)), :]
test_data = data[cv_set, :]
X_train = train_data[:, :-1]
Y_train = train_data[:, -1]
X_test = test_data[:, :-1]
Y_test = test_data[:, -1]
tree = DecisionTree()
tree.train(X_train, Y_train, cnt_cutoff, gain_cutoff)
pred_train = tree.run(X_train)
train_ac.append(accuracy(Y_train.T, pred_train.flatten()))
pred_test = tree.run(X_test)
test_ac.append(accuracy(Y_test.T, pred_test.flatten()))
tp += np.logical_and(Y_test.flatten(), pred_test.flatten()).sum()
tn += np.logical_and(np.logical_not(Y_test.flatten()),
np.logical_not(pred_test.flatten())).sum()
fp += np.logical_and(Y_test.flatten(),
np.logical_not(pred_test.flatten())).sum()
fn += np.logical_and(np.logical_not(Y_test.flatten()),
pred_test.flatten()).sum()