def trainAdaBoost(x, y, num_of_trees):
rows, _ = np.shape(x)
class_names = np.unique(y)
class_names = np.flip(class_names)
#Weights for each class in each 'parallell' tree
list_of_ensembles = [
] # Three ensembles for each class_name poor, median, excellent
for i in range(len(class_names)):
y_binary = convertOneVsAllToBinary(y, class_names[i])
ensemble = []
weights = np.ones(shape=(
rows,
1,
), dtype=float)
for _ in range(num_of_trees):
stump = Tree(x, y_binary, 1, weights)
stump.train(x, y_binary, 'boosting', weights)
y_pred = testTree(x, stump)
y_pred = y_pred.astype(int)
epsilon = calculateEpsilon(y_binary, y_pred, weights)
alpha = calculateAlpha(epsilon)
ensemble.append((
alpha,
stump,
))
#reweights
weights = reweight(y_binary, y_pred, weights)
# should make three ensembles
list_of_ensembles.append(ensemble)
return list_of_ensembles
def trainTree(x, y, depth, weights=None):
# return tree that have been trained
# Initiate tree object
dTree = Tree(x, y, depth) # Need to create the tree
dTree.train(x, y, 'boosting', weights) # Train it bruh
# TBD
return dTree
def trainBaggingEnsemble(x, y, depth, num_of_trees):
percentage = 3 / 4
ensemble = []
rows, _ = np.shape(x)
row_indexes = np.arange(rows)
num_of_training_data = math.ceil(rows * percentage)
for _ in range(num_of_trees):
# I want to only use 75% of the data when training
# Randomize which data to pull 75% out of x for each tree
np.random.shuffle(row_indexes)
x = x[row_indexes, :]
y = y[row_indexes, :]
training_x = x[0:num_of_training_data, :]
training_y = y[0:num_of_training_data, :]
dTree = Tree(training_x, training_y, depth)
dTree.train(training_x, training_y, 'bagging')
ensemble.append(dTree)
return ensemble