文档

Porting with Estimator

About Estimator

As a high-level API of TensorFlow 1.10 released in 2018, Estimator greatly streamlines the programming process of machine learning. Estimator has many advantages, for example, good support for distribution, simplified model creation, and code sharing between model developers.

TensorFlow 2.6 continues the support for this high-level API. To use it in the same way as in TensorFlow1.x, call it using the compat.v1 module as follows:

tf.compat.v1.estimator.Estimator

Develop your training script with the Estimator API of compat.v1 as follows:

  1. Data preprocessing: Create an input function input_fn.
  2. Model construction: Create a model function model_fn.
  3. Run configuration setting: Instantiate Estimator and pass the RunConfig object as the run parameter.
  4. Training: Call Estimator.train() to train your model with a fixed number of steps.

The following guides you through migrating your training script developed with Estimator, which after porting can run on the Ascend AI Processor.

Header File Inclusion

To import NPU-related libraries, add this header file reference in related Python files as follows:

1
2
3
import npu_device
from npu_device.compat.v1.npu_init import *
npu_device.compat.enable_v1()

Data Preprocessing

The code snippet is ready to use in normal cases. Manual tweaking is required only in the following scenario:

If the original network script relies on dataset.batch(batch_size) to return the dynamic shape, the shape of the last step on the network may be inconsistent with the previous shape because the number of remaining samples in the data flow may be less than the batch size. In this scenario, the dynamic shape compilation process starts. To improve network compilation performance, you are advised to set drop_remainder to True to discard the last several samples in the file and ensure that the shape of each step on the network is the same.
1
  dataset = dataset.batch(batch_size, drop_remainder=True)
Note that during inference, if the inference data volume of the last iteration is less than batch_size, you need to pad the inference data with blank data to batch_size. Failure to do so may lead to an assertion in your script that the number of validation results must be equal to the number of validation samples.
1
 assert num_written_lines == num_actual_predict_examples

Model Building

The code snippet is ready-to-use in normal cases. Manual tweaking is required only in the following scenarios:

  • If tf.device is used in the original network, delete the related code.
  • Replace gelu in the original network with the corresponding CANN API.
    Original TensorFlow code:
    1
2
3
4
5
    def gelu(x): 
  cdf = 0.5 * (1.0 + tf.tanh(
     (np.sqrt(2 / np.pi) * (x + 0.044715 * tf.pow(x, 3))))) 
  return x*cdf
layers = gelu()

    Code after porting:

    1
    layers = npu_unary_ops.gelu(x)

Run Configuration Setting

TensorFlow-based run parameters are configured through RunConfig. Modify the configuration as follows:

Original TensorFlow code:
1
2
3
4
5
6
7
8
session_config=tf.compat.v1.ConfigProto(allow_soft_placement=True,log_device_placement=False)
 
config=tf.estimator.RunConfig(
  session_config=session_config,
  model_dir=FLAGS.model_dir,  
  save_checkpoints_steps=FLAGS.save_checkpoints_steps, 
  ... ...
  )

Code after porting:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
session_config=tf.compat.v1.ConfigProto(allow_soft_placement=True,log_device_placement=False)
custom_op = sess_config.graph_options.rewrite_options.custom_optimizers.add()
custom_op.name = "NpuOptimizer"
sess_config.graph_options.rewrite_options.remapping = rewriter_config_pb2.RewriterConfig.OFF
 
npu_config=NPURunConfig(
  session_config=sess_config,
  model_dir=FLAGS.model_dir,  
  save_checkpoints_steps=FLAGS.save_checkpoints_steps,
  ... ...
  )

Estimator Creation

To port Estimator, simply change its config parameter to npu_config and port itself to NPUEstimator.

Original TensorFlow code:

1
2
3
4
mnist_classifier=tf.compat.v1.estimator.Estimator( 
  model_fn=cnn_model_fn, 
  config=config, 
  model_dir="/tmp/mnist_convnet_model")

Code after porting:

1
2
3
4
5
mnist_classifier=NPUEstimator( 
  model_fn=cnn_model_fn, 
  config=npu_config, 
  model_dir="/tmp/mnist_convnet_model" 
  )

Training

When training your model, specify the number of steps. The code snippet is ready-to-use in normal cases.
1
2
3
4
mnist_classifier.train(
  input_fn=train_input_fn,
  steps=20000,
  hooks=[logging_hook])