Description
Trainer component performs a training of Tensorflow model. The component is used only in creating Training pipelines. The Robotika Trainer container is the inheritor of the Trainer TFx [1] container with an extended UX.
Data source
- “URL of the training file” contains a path to the training file in the
- “Instance type” let you choose the the CPU/GPU configuration for the training.
- The requirements field contains a file with libraries.
- Framework version is the version of the Python.
How to use
There are two ways to load a training file into the Robotika component. The first way is to indicate a git path within a git repository that was intorduced in the first part of the creating the pipeline. Another way is to upload directly the file to the Robotika container.
The trainer script should contain a function with a name run_fn. The function takes the arguments used to train the model as name/value pairs. The function performs the training and stores the metadata on Robotika servers. The example of the run_fn is presented below: