TensorFlow 1.4: How to use BoostedTreesClassifier with Colab TPUs

Question

I have code for BoostedTreesClassifier which works, but takes a very long time with the amount of data i'm feeding it and the parameters i'm choosing i.e max_depth https://www.tensorflow.org/api_docs/python/tf/estimator/BoostedTreesClassifier

I'm attempting to use BoostedTreesClassifier estimator in Colab with TPUs, using TPUEstimator https://www.tensorflow.org/api_docs/python/tf/contrib/tpu/TPUEstimator

Is BoostedTreesClassifier possible with TPUEstimator? I am seeing that only neural networks can be used with Estimator/TPUEstimator https://www.tensorflow.org/guide/using_tpu

What is the correct approach to get BoostedTreesClassifier working with Colab TPUs?

tpu_estimator = tf.contrib.tpu.TPUEstimator(
    model_fn=model_fn,
    config=my_tpu_run_config,
    train_batch_size=100,
    use_tpu=True)

Answer 1

I think using the TPUStrategy is the right approach, but it's still taking a long time for some reason.

import pandas as pd
import numpy as np
import tensorflow as tf
print(tf.__version__)


resolver = tf.distribute.cluster_resolver.TPUClusterResolver()
tf.tpu.experimental.initialize_tpu_system(resolver)
tpu_strategy = tf.distribute.experimental.TPUStrategy(resolver)

with tpu_strategy.scope():
  model = tf.estimator.BoostedTreesClassifier(
      feature_columns=attibute_columns,
      n_batches_per_layer=10,
      center_bias=True,
      n_trees=100,
      max_depth=20,
      pruning_mode='post',
      tree_complexity=0.1)

model.train(input_fn=train_input_fn)
results = model.evaluate(eval_input_fn)
print(results)