def test_depth_tune_regression():
"""
test depth_tune, regression rf (structure check)
"""
n = 200
X = np.random.uniform(size=(n, 510), low=-1, high=1)
y = 10 * np.sin(np.pi * X[:,0]*X[:,1]) + 20 * ( X[:,2] - .5)**2 +\
10 * X[:,3] + 5 * X[:,4] + np.random.normal(size = n)
rf_class = sklearn.ensemble.RandomForestRegressor(n_estimators=2)
random_forest = rf_class.fit(X=X, y=y.ravel())
X_trained = X
y_trained = y
try:
new_rf = smooth_rf.depth_tune(random_forest,
X_trained,
y_trained,
verbose=False)
except:
assert False, \
"Error in tuning regression rf with oob depth"
assert type(new_rf) == type(random_forest), \
"updated random forest should be same class as random forest put in"
loss_vec = new_rf.loss_vec_depth
_, max_depth = smooth_rf.calc_depth_for_forest(random_forest,
verbose=False)
assert loss_vec.shape[0] == np.int(max_depth), \
"loss vector is incorrect dimension relative to maximum depth of rf"
def test_depth_tune_classification():
"""
test depth_tune, classification rf (structure check)
"""
n = 200
min_size_leaf = 1
X = np.random.uniform(size=(n, 510), low=-1, high=1)
y = 10 * np.sin(np.pi * X[:,0]*X[:,1]) + 20 * ( X[:,2] - .5)**2 +\
10 * X[:,3] + 5 * X[:,4] + np.random.normal(size = n)
y_cat = np.array(pd.cut(y, bins=5, labels=np.arange(5, dtype=np.int)),
dtype=np.int)
y = y_cat
num_classes = len(Counter(y_cat).keys())
rf_class = sklearn.ensemble.RandomForestClassifier(
n_estimators=2, min_samples_leaf=min_size_leaf)
random_forest = rf_class.fit(X=X, y=y.ravel())
try:
new_rf = smooth_rf.depth_tune(random_forest, X, y, verbose=False)
except:
assert False, \
"Error in tuning regression rf with oob depth"
assert type(new_rf) == type(random_forest), \
"updated random forest should be same class as random forest put in"
loss_vec = new_rf.loss_vec_depth
_, max_depth = smooth_rf.calc_depth_for_forest(random_forest,
verbose=False)
assert loss_vec.shape[0] == np.int(max_depth), \
"loss vector is incorrect dimension relative to maximum depth of rf"
## models ##
# base rf --------------
rf_base = generate_rf(X_train, y_train, n_trees, reg_or_class=reg_or_class)
info_dict["seed"] = my_seed
scoring_dict["rf_base"] = assess_rf(rf_base, X_test, y_test)
random_forest = copy.deepcopy(rf_base)
# depth analysis --------------
print("Depth tune:")
time_start = time.time()
depth_tune_rf = smooth_rf.depth_tune(random_forest,
X_trained=X_train,
y_trained=y_train)
scoring_dict["depth_tune"] = assess_rf(depth_tune_rf, X_test, y_test)
info_dict["depth_tune"] = depth_tune_rf.loss_vec_depth
depth_spent = time.time() - time_start
times = [depth_spent]
# wOLS analysis --------------
if reg_or_class == "reg":
print("wOLS:")
time_start = time.time()
parent_all_opts = [True, False] # parent_all
no_constraint_opts = [True, False] # no_constraint