Keras Train¶
Overview¶
Compatible Pyspark only
Python node : launch with punchplatform-pyspark.sh or the punch UI job-editor
with environment pyspark
.
This node define and train a small demonstrative keras model to recognise digits from the mnist database. It publish the trained keras model to the output stream.
The model created is a feed-forward neural network with one hidden layer of 128 neurons. The training is done over 5 epoch with a batch size of 50.
Since it's mostly a specific demo node, it don't have any settings.
Example configuration :¶
1 2 3 4 5 6 7 8 9 | { type: keras_train component: trainer publish: [ { stream: model } ] } |
Example of complete pipeline :¶
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | { type: pyspark job: [ { type: keras_train component: trainer publish: [ { stream: model } ] } { type: keras_save_model component: model_save settings:{ path: my_model.h5 } subscribe:[ { component: trainer stream: model } ] } ] } |