label = {'y': ['yes', 'no']}
num_run = 100
T = 1000
test_py = np.array([[0 for x in range(test_size)] for y in range(num_run)])
test_py_first = np.array([0 for x in range(test_size)])
for iter in range(num_run):
train_subset = train_data.sample(n=1000, replace=False, random_state=iter)
for t in range(T):
print('iter: ', iter, 't: ', t)
# sample with replace
sampled = train_subset.sample(frac=0.01, replace=True, random_state=t)
# ID3
dt_generator = dt.ID3(feature_selection=0, max_depth=17, subset=6)
# get decision tree
decision_tree = dt_generator.generate_decision_tree(
sampled, features, label)
## predict
# test
py = dt_generator.classify(decision_tree, test_data)
py = np.array(py.tolist())
py[py == 'yes'] = 1
py[py == 'no'] = -1
py = py.astype(int)
test_py[iter] = test_py[iter] + py
if t == 0:
test_py_first = test_py_first + py
true_value = np.array(test_data['y'].tolist())
'maint': ['vhigh', 'high', 'med', 'low'],
'doors': ['2', '3', '4', '5more'],
'persons': ['2', '4', 'more'],
'lug_boot': ['small', 'med', 'big'],
'safety': ['low', 'med', 'high']
}
label = {'label': ['unacc', 'acc', 'good', 'vgood']}
train_acc = [[0 for x in range(6)] for y in range(3)]
test_acc = [[0 for x in range(6)] for y in range(3)]
for feature_selection in range(3):
for max_depth in range(6):
# ID3
dt_generator = dt.ID3(feature_selection=feature_selection,
max_depth=max_depth + 1)
# get decision tree
decision_tree = dt_generator.generate_decision_tree(
train_data, features, label)
# train acc
# predict
train_data['plabel'] = dt_generator.classify(decision_tree, train_data)
train_acc[feature_selection][max_depth] = train_data.apply(
lambda row: 1
if row['label'] == row['plabel'] else 0, axis=1).sum() / train_size
# test acc
# predict
test_data['plabel'] = dt_generator.classify(decision_tree, test_data)
test_acc[feature_selection][max_depth] = test_data.apply(
lambda row: 1
if row['label'] == row['plabel'] else 0, axis=1).sum() / test_size