Tensorflow DNNRegressor Multiple Outputs

user_8302055 0
Viewed 1119

I am trying to use tf.contrib.learn.DNNRegressor to model a multi-input multi-output system. I have followed the Boston DNNRegressor example on the Tensorflow website, however when I try to pass an array of 2 outputs to the regressor fitter, I get

raise ValueError("Shapes %s and %s are incompatible" % (self, other))
ValueError: Shapes (100, 1) and (100, 2) are incompatible

I found this post with no responses from back in January, so it seems other people have had this problem.

I could conceivably use multiple DNNRegressors for each of the outputs, however is it possible to predict multiple outputs with a single DNNRegressor in Tensorflow?

I am running Tensorflow 1.2.1 on Ubuntu 16.04.

tensorflow machine-learning python
1 Answers

This might be more than a little too late, but the current tf.estimator.DNNRegressor has an argument label_dimension that might do what you're looking for.

regressor = estimator.DNNRegressor(feature_columns=my_feature_columns,
                                   label_dimension=2,
                                   hidden_units=hidden_layers,
                                   model_dir=MODEL_PATH)

In my case this deals perfectly with two outputs

user_58369950
Related