MXNet sequential regression model example

An example of a sequential network that solves a regression task used as an OpenML flow.

import mxnet
import mxnet.gluon

import openml
import openml.extensions.mxnet

import logging

Enable logging in order to observe the progress while running the example.

openml.config.logger.setLevel(logging.DEBUG)
openml.extensions.mxnet.config.logger.setLevel(logging.DEBUG)

Define a HybridSequential container

model = mxnet.gluon.nn.HybridSequential()

Add the layers to the HybridSequential container

with model.name_scope():
    model.add(mxnet.gluon.nn.BatchNorm())
    model.add(mxnet.gluon.nn.Dense(units=1024, activation="relu"))
    model.add(mxnet.gluon.nn.Dropout(rate=0.4))
    model.add(mxnet.gluon.nn.Dense(units=512, activation="relu"))
    model.add(mxnet.gluon.nn.Dropout(rate=0.4))
    model.add(mxnet.gluon.nn.Dense(units=256, activation="relu"))
    model.add(mxnet.gluon.nn.Dropout(rate=0.4))
    model.add(mxnet.gluon.nn.Dense(units=128, activation="relu"))
    model.add(mxnet.gluon.nn.Dropout(rate=0.4))
    model.add(mxnet.gluon.nn.Dense(units=64, activation="relu"))
    model.add(mxnet.gluon.nn.Dropout(rate=0.4))
    model.add(mxnet.gluon.nn.Dense(units=1, activation="relu"))

Enable hybrid execution

model.hybridize()

Retrieve the credit_g classification task from OpenML

task = openml.tasks.get_task(2295)

Run the model on the task (requires an API key).

run = openml.runs.run_model_on_task(model, task, avoid_duplicate_runs=False)
# Publish the run on OpenML
run.publish()

print('URL for run: %s/run/%d' % (openml.config.server, run.run_id))