robieta set strip_default_attrs=True for SavedModel exports (#5439)
* set strip_default_attrs=True for SavedModel exports

* specify dtype in resnet export

* another dtype fix

* fix another dtype issue, and set --image_bytes_as_serving_input to default to False
Latest commit cdcd3ec Oct 5, 2018

MNIST in TensorFlow

This directory builds a convolutional neural net to classify the MNIST dataset using the, tf.estimator.Estimator, and tf.layers APIs.


To begin, you'll simply need the latest version of TensorFlow installed. First make sure you've added the models folder to your Python path; otherwise you may encounter an error like ImportError: No module named official.mnist.

Then to train the model, run the following:


The model will begin training and will automatically evaluate itself on the validation data.

Illustrative unit tests and benchmarks can be run with:

python --benchmarks=.

Exporting the model

You can export the model into Tensorflow SavedModel format by using the argument --export_dir:

python --export_dir /tmp/mnist_saved_model

The SavedModel will be saved in a timestamped directory under /tmp/mnist_saved_model/ (e.g. /tmp/mnist_saved_model/1513630966/).

Getting predictions with SavedModel Use saved_model_cli to inspect and execute the SavedModel.

saved_model_cli run --dir /tmp/mnist_saved_model/TIMESTAMP --tag_set serve --signature_def classify --inputs image=examples.npy

examples.npy contains the data from example5.png and example3.png in a numpy array, in that order. The array values are normalized to values between 0 and 1.

The output should look similar to below:

Result for output key classes:
[5 3]
Result for output key probabilities:
[[  1.53558474e-07   1.95694142e-13   1.31193523e-09   5.47467265e-03
    5.85711526e-22   9.94520664e-01   3.48423509e-06   2.65365645e-17
    9.78631419e-07   3.15522470e-08]
 [  1.22413359e-04   5.87615965e-08   1.72251271e-06   9.39960718e-01
    3.30306928e-11   2.87386645e-02   2.82353517e-02   8.21146413e-18
    2.52568233e-03   4.15460236e-04]]

Experimental: Eager Execution

Eager execution (an preview feature in TensorFlow 1.5) is an imperative interface to TensorFlow. The exact same model defined in can be trained without creating a TensorFlow graph using:


Experimental: TPU Acceleration (and demonstrate training a neural network to classify digits on CPUs and GPUs. can be used to train the same model using TPUs for hardware acceleration. More information in the tensorflow/tpu repository.