def run_naive_rerf(dataset_name,
data,
choosen_classes,
sub_train_indices,
rf_type="shared"):
(train_images, train_labels), (test_images, test_labels) = get_subset_data(
dataset_name, data, choosen_classes, sub_train_indices)
time_taken = dict()
# Train
train_start = time.time()
forest = fastRerF(X=train_images.reshape(len(train_images), -1),
Y=train_labels,
forestType=TREE_TYPE,
trees=100,
numCores=cpu_count() - 1)
train_end = time.time()
time_taken["train"] = train_end - train_start
# forest.printParameters()
# Test
test_start = time.time()
test_preds = fastPredict(test_images.reshape(len(test_images), -1), forest)
test_end = time.time()
time_taken["test"] = test_end - test_start
return accuracy_score(test_labels, test_preds), time_taken
def run_one_layer_deep_conv_rf(dataset_name,
data,
choosen_classes,
sub_train_indices,
type="shared"):
(train_images, train_labels), (test_images, test_labels) = get_subset_data(
dataset_name, data, choosen_classes, sub_train_indices)
time_taken = dict()
# ConvRF (layer 1)
if type == "rerf_shared":
conv1 = DeepConvRF(type="rerf_shared",
kernel_size=10,
stride=2,
rerf_params={
"num_trees": RERF_NUM_TREES,
"tree_type": RERF_TREE_TYPE
})
else:
conv1 = DeepConvRF(type=type, kernel_size=10, stride=2)
conv1_map = conv1.convolve_fit(train_images, train_labels)
conv1_map_test = conv1.convolve_predict(test_images)
time_taken = copy.deepcopy(conv1.time_taken)
# Full RF
train_start = time.time()
if type == "rerf_shared":
conv1_full_RF = fastRerF(X=conv1_map.reshape(len(train_images), -1),
Y=train_labels,
forestType=RERF_TREE_TYPE,
trees=100,
numCores=cpu_count() - 1)
else:
conv1_full_RF = RandomForestClassifier(n_estimators=100, n_jobs=-1)
conv1_full_RF.fit(conv1_map.reshape(len(train_images), -1),
train_labels)
train_end = time.time()
time_taken["train"] += (train_end - train_start)
time_taken["final_fit"] = (train_end - train_start)
test_start = time.time()
if type == "rerf_shared":
test_preds = fastPredict(conv1_map_test.reshape(len(test_images), -1),
conv1_full_RF)
else:
test_preds = conv1_full_RF.predict(
conv1_map_test.reshape(len(test_images), -1))
test_end = time.time()
time_taken["test"] += (test_end - test_start)
time_taken["final_predict"] = (test_end - test_start)
return accuracy_score(test_labels, test_preds), time_taken
# CSVFile=datafile,
# Ycolumn=label_col,
# forestType="binnedBaseRerF",
# trees=500,
# numCores=cpu_count(),
# )
forest = fastRerF(X=feat_data,
Y=labels,
forestType="binnedBaseRerF",
trees=500,
numCores=cpu_count())
forest.printParameters()
# training predictions
predictions = fastPredict(feat_data, forest)
# print(predictions)
# training posterior predictions probabilities
post_pred = fastPredictPost(feat_data, forest)
# print(post_pred)
print("Error rate", np.mean(predictions != labels))
print("loading test data...")
if datatype == "iris":
data_fname = "./iris.csv" # iris
elif datatype == "mnist":
data_fname = "../packedForest/res/mnist_test.csv" # mnist
test_X = np.genfromtxt(data_fname, delimiter=",")