def train_svm_forest(X, Y, num_trees = 10, max_depth = 3, bagging_percent=0.65, randomize_C = False, model_args ={}, tree_args={}):
"""A random forest whose base classifier is a SVM-Tree (rather
than splitting individual features we project each point onto a hyperplane)
Parameters
----------
X : numpy array containing input data.
Should have samples for rows and features for columns.
Y : numpy array containing class labels for each sample
num_trees : how big is the forest?
bagging_percent : what subset of the data is each tree trained on?
randomize_C : bool
model_args : parameters for each SVM classifier
tree_args : parameters for each tree of classifiers
"""
tree = mk_svm_tree(max_depth, randomize_C, model_args, tree_args)
forest = ClassifierEnsemble(
base_model = tree,
num_models = num_trees,
bagging_percent = bagging_percent)
forest.fit(X,Y)
return forest
def train_sgd_forest(X, Y,
num_trees = 20,
max_depth = 3,
bagging_percent=0.65,
randomize_alpha=False,
model_args = {},
tree_args= {}):
"""A random forest whose base classifier is a tree of SGD classifiers
Parameters
----------
X : numpy array containing input data.
Should have samples for rows and features for columns.
Y : numpy array containing class labels for each sample
num_trees : how big is the forest?
bagging_percent : what subset of the data is each tree trained on?
randomize_alpha : bool
model_args : parameters for each SGD classifier
tree_args : parameters for each tree
"""
bagsize = bagging_percent * X.shape[0]
tree = mk_sgd_tree(bagsize, max_depth, randomize_alpha, model_args, tree_args)
forest = ClassifierEnsemble(
base_model = tree,
num_models = num_trees,
bagging_percent = bagging_percent)
forest.fit(X,Y)
return forest
def test_stacked_random_forest():
t = RandomizedTree(min_leaf_size=1)
lr = LogisticRegression()
ensemble = ClassifierEnsemble(base_model=t, stacking_model=lr)
ensemble.fit(data, labels)
try_predictor(ensemble)
def test_stacked_random_forest():
t = RandomizedTree(min_leaf_size=1)
lr = LogisticRegression()
ensemble = ClassifierEnsemble(base_model=t, stacking_model=lr)
ensemble.fit(data, labels)
try_predictor(ensemble)