Get Started with Secure XGBoost

This is a quick start tutorial showing snippets for you to quickly try out Secure XGBoost on the demo dataset on a binary classification task. This sanity check ensures that setup was done properly. This quickstart uses encrypted versions of the agaricus.txt.train and agaricus.txt.test data files from demo/data/.

Encrypt data

Secure XGBoost currently only supports data ingestion from text files. It supports two file formats: LibSVM and CSV files. See the original XGBoost documentation for more details on the input format for these file types, here.

The following snippet shows how clients can generate a symmetric key and use it to encrypt their data files.

import securexgboost as xgb

KEY_FILE = "key.txt"
xgb.generate_client_key(KEY_FILE)
xgb.encrypt_file("demo/data/agaricus.txt.train", "demo/data/train.enc", KEY_FILE)
xgb.encrypt_file("demo/data/agaricus.txt.test", "demo/data/test.enc", KEY_FILE)

Train a model on the encrypted data

Below is a snippet of the full Python demo located at secure-xgboost/demo/python/basic/secure-xgboost-demo.py. This demo runs the enclave server on the same machine as the client for simplicity.

Note

If you built Secure XGBoost in simulation mode, remote attestation will not work, as the simulated enclave will not generate a report. Consequently, report verification will not work, and you should set verify=False when calling attest().

import securexgboost as xgb

# Initialize client and connect to enclave
xgb.init_client(user_name="user1",
                sym_key_file="demo/data/key_zeros.txt",
                priv_key_file="config/user1.pem",
                cert_file="config/user1.crt")
xgb.init_server(enclave_image="build/enclave/xgboost_enclave.signed", client_list=["user1"])
# Remote Attestation
xgb.attest()

# Load the encrypted data remotely
dtrain = xgb.DMatrix({"user1": "demo/data/agaricus.txt.train.enc"})
dtest = xgb.DMatrix({"user1": "demo/data/agaricus.txt.test.enc"})

params = {
   "objective": "binary:logistic",
   "gamma": "0.1",
   "max_depth": "3"
}

# Train a model remotely
num_rounds = 5
booster = xgb.train(params, dtrain, num_rounds, evals=[(dtrain, "train"), (dtest, "test")])

# Get encrypted predictions and decrypt them locally
predictions, num_preds = booster.predict(dtest)