TensorFlow on jeffcarp

Things I Learned as a First-Time Intern Host

Sat, 31 Aug 2019 11:13:25 -0700

I hosted an intern for the first time this summer. It was my first time being somebody’s manager and it became a huge learning experience for me as well as a really fun time. My intern worked on adding many features to velocity-tracking charts, rewriting our ML models in TensorFlow 2.0, and a few other projects.

Here are the biggest areas where I struggled as a host and the important lessons I took away from those experiences.

Doing Cryptography in TensorFlow

Sat, 23 Jun 2018 13:00:14 -0700

On the left: the Feistel Network from the DES cipher, implemented below. On the right: a deep neural network.

TensorFlow is a popular machine learning framework. If you look under the hood, TensorFlow is a general platform for doing computation over tensors in the structure of a graph.

How to Export Evaluation Results in Tensorflow

Fri, 05 Jan 2018 21:33:26 -0700

2019 update: just a heads up, this post is about TensorFlow 1.x. For TensorFlow 2.x, you probably want to check out Keras custom callbacks.

In TensorFlow if you’re using a tf.estimator model, for instance tf.estimator.DNNLinearCombinedClassifier, and as part of your automated training infrastructure you want to save the evaluation results as a JSON file, it’s not super straightforward, so here’s how to do it.

Let’s say you define your EvalSpec like this:

eval_spec = tf.estimator.EvalSpec(eval_input_fn,
  steps=hparams.eval_steps,
  exporters=[exporter],
  name='eval')

You’ll need to write a new exporter class that will take the eval_result from your evaluation step and save it to a file using the GFile API.

Example: Save and Load a TensorFlow Model

Sun, 19 Nov 2017 19:07:30 +0000

2020 update: just a heads up, this post is about TensorFlow 1.x. For TensorFlow 2.x, you probably want to check out this guide .

This post details how to save and load a TensorFlow model using the DNNClassifier API.

The key idea here is that you define a function or a class beforehand that takes a model directory (in which it will save and restore the model parameters), adds that to RunConfig, and returns a tf.contrib.learn.Estimator, for example, tf.contrib.learn.DNNClassifier. See make_estimator for more details.