Kevin Kuo

What's your setup? I've seen that but have trouble getting a consistent reprex.

Possibly related: https://github.com/dmlc/xgboost/issues/2780

Turns out the published jars don't support windows, so we'd have to work around it. See the note at https://xgboost.readthedocs.io/en/latest/jvm/#access-snapshot-version. We could try to download from criteo-forks/xgboost-jars on the fly.

Looks like a bug! Do the multiclass (>2) models behave properly?

@wlandau Do you mind opening an issue to make sure we track it? That's something I wondered about, too.

More informative errors would be nice here!

Aren't we actually calling the [tf.estimator](https://github.com/tensorflow/tensorflow/blob/562045bb93743b09f2652924fdf300fc28fca3db/tensorflow/python/estimator/canned/linear.py) functions instead of the ones from tf.contrib.learn? The model functions should be returning the appropriate `export_outputs` dict e.g. at https://github.com/tensorflow/tensorflow/blob/9bdb72e124e50e1b12b3286b38cbb1c971552741/tensorflow/python/estimator/canned/head.py#L660 Not sure why `WARNING:tensorflow:Export...

@javierluraschi yeah `linear_regressor` is a [canned estimator](https://github.com/rstudio/tfestimators/blob/77f406c743ad59c14eda1f460ab6882e9cd6d2a3/R/linear_estimators.R#L31-L39) (in fact all estimators in `tfestimators` are canned except when you define your own via `tfestimators::estimator()`) and we hit `canned/head.py` [here](https://github.com/tensorflow/tensorflow/blob/r1.4/tensorflow/python/estimator/canned/linear.py#L337).

Do keras and tensorflow work properly otherwise?

Yeah agreed this is not beginner/newcomer friendly at all and we can do much better...