unit test

1. unit test

use unittest.TestCase or pytest or nose.

For the tensorflow based test, there is a `tf.test`.

--> for tensorflow:

Data

• Ensure that our data has the right format (yes I put it again here for completion)

• Ensure that the training labels are correct

• Test our complex processing steps such as image manipulation

• Assert data completion, quality, and errors

• Test the distribution of the features

Training

• Run a training step and compare the weight before and after to ensure that they are updated

• Check that our loss function can be actually used on our data

Evaluation:

• Having tests to ensure that your metrics ( e.g accuracy, precision, and recall ) are above a threshold when iterating over different architectures

• You can run speed/benchmark tests on training to catch possible overfitting

• Of course, cross-validation can be in the form of a unit test

Model Architecture:

• The model’s layers are actually stacking

• The model’s output has the correct shape

2. mocking

Mocking makes it very easy to replace complex logic or heavy dependencies when testing code using dummy objects.

3. coverage