How to use DPGradientDescentGaussianOptimizer for BoostedTreesClassifier?

[FelixNeutatz]

Dear all,

I am really excited about this library. I wanted to try it out with a standard model, such as BoostedTreesClassifier. Is there any way to change the underlying optimizer to a differentially private one?

Here, would be an example, where I would like to apply DP optimization:

import pandas as pd
import tensorflow as tf

from tensorflow_privacy.privacy.analysis import compute_dp_sgd_privacy_lib
from tensorflow_privacy.privacy.optimizers import dp_optimizer

dftrain = pd.read_csv('https://storage.googleapis.com/tf-datasets/titanic/train.csv')
dfeval = pd.read_csv('https://storage.googleapis.com/tf-datasets/titanic/eval.csv')
y_train = dftrain.pop('survived')
y_eval = dfeval.pop('survived')

fc = tf.feature_column
CATEGORICAL_COLUMNS = ['sex', 'n_siblings_spouses', 'parch', 'class', 'deck',
                       'embark_town', 'alone']
NUMERIC_COLUMNS = ['age', 'fare']


def one_hot_cat_column(feature_name, vocab):
    return tf.feature_column.indicator_column(
        tf.feature_column.categorical_column_with_vocabulary_list(feature_name, vocab))

feature_columns = []
for feature_name in CATEGORICAL_COLUMNS:
    vocabulary = dftrain[feature_name].unique()
    feature_columns.append(one_hot_cat_column(feature_name, vocabulary))

for feature_name in NUMERIC_COLUMNS:
    feature_columns.append(tf.feature_column.numeric_column(feature_name, dtype=tf.float32))

example = dict(dftrain.head(1))
class_fc = tf.feature_column.indicator_column(tf.feature_column.categorical_column_with_vocabulary_list('class', ('First', 'Second', 'Third')))

NUM_EXAMPLES = len(y_train)

def make_input_fn(X, y, n_epochs=None, shuffle=True):
  def input_fn():
    dataset = tf.data.Dataset.from_tensor_slices((dict(X), y))
    if shuffle:
      dataset = dataset.shuffle(NUM_EXAMPLES)
    dataset = dataset.repeat(n_epochs)
    dataset = dataset.batch(NUM_EXAMPLES)
    return dataset
  return input_fn

train_input_fn = make_input_fn(dftrain, y_train)
eval_input_fn = make_input_fn(dfeval, y_eval, shuffle=False, n_epochs=1)

n_batches = 1
est = tf.estimator.BoostedTreesClassifier(feature_columns, n_batches_per_layer=n_batches)
est.train(train_input_fn, max_steps=100)
result = est.evaluate(eval_input_fn)


print(result)

eps, _ = compute_dp_sgd_privacy_lib.compute_dp_sgd_privacy(60000, 256, 1.1, 30, 1e-5)
print('For delta=1e-5, the current epsilon is: %.2f' % eps)

Thank you for your help.

Best regards, Felix