def test_rf(n, reps, n_estimators):
preds = np.zeros((reps, 10000))
acc = np.zeros(reps)
for i in range(reps):
X_train, y_train, X_test, y_test = load_data(n)
clf = rfc(n_estimators=n_estimators, projection_matrix="Base")
clf.fit(X_train, y_train)
preds[i] = clf.predict(X_test)
acc[i] = np.sum(preds[i] == y_test) / len(y_test)
np.save("output/rf_orthant_preds_" + str(n) + ".npy", preds)
return acc
feature_combinations = [2, 3, 4, 5]
max_features = [3, 5, 10, 20] # ceil of 20 ^ (0.25, 0.5, 0.75, 1)
dftrain = np.load("data/orthant_train_400.npy")
dftest = np.load("data/orthant_test.npy")
X_train = dftrain[:, :-1]
y_train = dftrain[:, -1]
X_test = dftest[:, :-1]
y_test = dftest[:, -1]
param_acc = np.zeros((4, 4))
for i, f in enumerate(feature_combinations):
for j, m in enumerate(max_features):
for k in range(3):
clf = rfc(n_estimators=n_estimators,
projection_matrix="RerF",
feature_combinations=f,
max_features=m)
clf.fit(X_train, y_train)
preds = clf.predict(X_test)
param_acc[i, j] = np.sum(preds == y_test) / len(y_test)
print(param_acc)
np.save("orthant_gridsearch", param_acc)
from sklearn.datasets import load_iris
from rerf.rerfClassifier import rerfClassifier as rfc
from proglearn.transformers import ObliqueTreeClassifier as of
data = load_iris()
X, y = data.data, data.target
rerf_acc = np.zeros(5)
of_acc = np.zeros(5)
for i in range(5):
OF = of( #oblique=True,
max_features=1.0, feature_combinations=1.0)
RFC = rfc(n_estimators=1,
projection_matrix="RerF",
feature_combinations=1.0,
max_features=1.0)
# Rerf
RFC.fit(X, y)
rerf_preds = RFC.predict(X)
rerf_acc[i] = np.sum(rerf_preds == y) / len(y)
# Of
OF.fit(X, y)
of_preds = OF.predict(X)
of_acc[i] = np.sum(of_preds == y) / len(y)
print("RerF: ", np.mean(rerf_acc))
print("OF: ", np.mean(of_acc))