def CreateFullNewTree(dtree):
d = dict()
d['node_count'] = 1
d['max_depth'] = 0
d['nodes'] = np.zeros(1,
dtype=[('left_child', '<i8'), ('right_child', '<i8'),
('feature', '<i8'), ('threshold', '<f8'),
('impurity', '<f8'),
('n_node_samples', '<i8'),
('weighted_n_node_samples', '<f8')])
d['values'] = np.zeros(
(1, dtree.tree_.value.shape[1], dtree.tree_.value.shape[2]))
(Tree, (n_f, n_c, n_o), b) = dtree.tree_.__reduce__()
new_tree = Tree(n_f, n_c, n_o)
new_tree.__setstate__(d)
new_dtree = DecisionTreeClassifier()
new_dtree.n_features_ = n_f
new_dtree.n_classes_ = n_c[0]
new_dtree.classes_ = np.linspace(0, n_c[0] - 1, n_c[0]).astype(int)
new_dtree.n_outputs_ = n_o
new_dtree.tree_ = new_tree
return new_dtree
def build_decision_tree(t):
dt = DecisionTreeClassifier(random_state=0)
dt.n_features_ = t.n_features
dt.n_outputs_ = t.n_outputs
dt.n_classes_ = t.n_classes[0]
dt.classes_ = np.array([x for x in range(dt.n_classes_)])
dt.tree_ = t
return dt
def equivalent_random(dtree):
dic_or = dtree.tree_.__getstate__().copy()
leaves = np.where(dtree.tree_.feature == -2)[0]
all_splits = np.zeros(dtree.tree_.node_count - leaves.size, dtype=object)
compt = 0
for i in range(dtree.tree_.node_count):
if i not in leaves:
all_splits[compt] = (dtree.tree_.feature[i],
dtree.tree_.threshold[i])
compt = compt + 1
print(all_splits)
d = dict()
d['node_count'] = 1
d['max_depth'] = 0
d['nodes'] = np.zeros(1,
dtype=[('left_child', '<i8'), ('right_child', '<i8'),
('feature', '<i8'), ('threshold', '<f8'),
('impurity', '<f8'),
('n_node_samples', '<i8'),
('weighted_n_node_samples', '<f8')])
d['values'] = np.zeros(
(1, dic_or['values'].shape[1], dic_or['values'].shape[2]))
#initial node
k = np.random.randint(all_splits.size)
phi_init, th_init = all_splits[k]
d['nodes'][0]['feature'] = phi_init
d['nodes'][0]['threshold'] = th_init
d['nodes'][0]['left_child'] = -1
d['nodes'][0]['right_child'] = -1
d = rec_split(dic_or, 0, d, all_splits)
print('nb noeuds :', d['node_count'])
(Tree, (n_f, n_c, n_o), b) = dtree.tree_.__reduce__()
new_tree = Tree(n_f, n_c, n_o)
new_tree.__setstate__(d)
print('nb noeuds :', new_tree.__getstate__()['nodes'].size)
new_dtree = DecisionTreeClassifier()
new_dtree.n_features_ = n_f
new_dtree.n_classes_ = n_c[0]
#print(n_c)
new_dtree.classes_ = np.linspace(0, n_c[0] - 1, n_c[0]).astype(int)
new_dtree.n_outputs_ = n_o
new_dtree.tree_ = new_tree
new_dtree.max_depth = new_tree.max_depth
return new_dtree
def deserialize_decision_tree(model_dict):
deserialized_model = DecisionTreeClassifier(**model_dict['params'])
deserialized_model.classes_ = np.array(model_dict['classes_'])
deserialized_model.max_features_ = model_dict['max_features_']
deserialized_model.n_classes_ = model_dict['n_classes_']
deserialized_model.n_features_ = model_dict['n_features_']
deserialized_model.n_outputs_ = model_dict['n_outputs_']
tree = deserialize_tree(model_dict['tree_'], model_dict['n_features_'], model_dict['n_classes_'], model_dict['n_outputs_'])
deserialized_model.tree_ = tree
return deserialized_model
def train_decision_tree_classifer(training_data, depth=50):
dtree = DecisionTreeClassifier(max_depth=depth,
min_samples_split=2,
class_weight="balanced")
dtree.classes_ = [0, 1]
scores = cross_val_score(dtree,
training_data[:,
selected_features].astype('float'),
training_data[:, -1].astype('float'),
cv=5,
scoring='roc_auc')
print("Scores gotten using Decision Tree (max depth=" + str(depth) + ")")
print(scores)
print(np.mean(scores))
return dtree, np.mean(scores)
def fed_integrate_model_cart(model1, model2):
coef_1 = model1.coef_
coef_2 = model2.coef_
intercept_1 = model1.intercept_
intercept_2 = model2.intercept_
classes = model1.classes_
model = DecisionTreeClassifier()
model.coef_ = mulkeys_add_2Darray(model1.coef_, model2.coef_)/2
model.intercept_ = mulkeys_add_array(
model1.intercept_, model1.intercept_)/2
model.classes_ = classes
return model
