This tutorial shows how to build a simple CNN on the MNIST dataset using Determined's tf.keras API. This example is adapted from this Keras image classification tutorial.
- model_def.py: The core code for the model. This includes building and compiling the model.
- data.py: The data loading and preparation code for the model.
- const.yaml: Train the model with constant hyperparameter values.
- distributed.yaml: Same as
const.yaml
, but trains the model with multiple GPUs (distributed training). - adaptive.yaml: Perform a hyperparameter search using Determined's state-of-the-art adaptive hyperparameter tuning algorithm.
The current implementation downloads the Fashion MNIST data from here.
If you have not yet installed Determined, installation instructions can be found
under docs/install-admin.html
or at https://docs.determined.ai/latest/index.html
Run the following command: det -m <master host:port> experiment create -f const.yaml .
. The other configurations can be run by specifying the appropriate
configuration file in place of const.yaml
.
Training the model with the hyperparameter settings in const.yaml
should yield
a validation accuracy of ~85%.