Skip to main content

Keras with TensorBoard

The example below demonstrates the integration of ClearML into code which uses Keras and TensorBoard. View it in script or in Jupyter Notebook.

note

The example in Jupyter Notebook includes a clickable icon to open the notebook in Google Colab.

The example script does the following:

  1. Trains a simple deep neural network on the Keras built-in MNIST dataset.
  2. Builds a sequential model using a categorical cross entropy loss objective function.
  3. Specifies accuracy as the metric, and uses two callbacks: a TensorBoard callback and a model checkpoint callback.
  4. During script execution, creates an experiment named Keras with TensorBoard example in the examples project (in script) or the Colab notebooks project (in Jupyter Notebook).

Scalars​

The loss and accuracy metric scalar plots appear in SCALARS, along with the resource utilization plots, which are titled :monitor: machine.

image

Histograms​

Histograms for layer density appear in PLOTS.

image

Hyperparameters​

ClearML automatically logs command line options generated with argparse, and TensorFlow Definitions.

Command line options appear in CONFIGURATION > HYPERPARAMETERS > Args.

image

TensorFlow Definitions appear in TF_DEFINE.

image

Console​

Text printed to the console for training progress, as well as all other console output, appear in CONSOLE.

image

Configuration Objects​

In the experiment code, a configuration dictionary is connected to the Task by calling Task.connect(). 

task.connect_configuration(
   name="MyConfig", 
   configuration={'test': 1337, 'nested': {'key': 'value', 'number': 1}}
)

It appears in CONFIGURATION > CONFIGURATION OBJECTS > MyConfig.

image