Travis Addair

Hey @fire, this should be fixed now in v0.5.5. Please try it out and let us know how it goes!

Hey @Karmadon, one option for you would be to export your Ludwig model to a [Neuropod](https://github.com/uber/neuropod), which we have an API for [here](https://github.com/ludwig-ai/ludwig/blob/master/ludwig/export.py#L59). Once exported to Neuropod, you can use...

@vkuzmin-uber @VivekPanyam are there still plans to add Golang bindings to Neuropod? I see there's this issue that is still open: https://github.com/uber/neuropod/issues/294 Any updates?

Config: ``` Experiment name: BERT_128_unweighted_4jan_712 Model name: bert_128 Output directory: models/results/BERT_128_unweighted_4jan_712_bert_128_0 ludwig_version: '0.4' command: ('/usr/local/lib/python3.8/dist-packages/ipykernel_launcher.py -f ' '/root/.local/share/jupyter/runtime/kernel-8a7019b2-4606-4285-b569-e4be5dc2edc8.json') random_seed: 42 data_format: "" config: { 'combiner': {'type': 'concat'}, 'input_features': [ {...

@smiraldr can you additional context from our Slack conversation here?

Hi @Jeffwan, I think the approach of using a Job should work well, particularly to support retires in the event of node failure. The [docs](https://github.com/ludwig-ai/ludwig/tree/master/examples/ray/kubernetes) we have for running with...

Different proposals: ``` # 1. This LudwigConfig(input_features=[ BinaryInputFeature(name="foo", encoder=Dense(num_layers=3), preprocessing=...) ]) # 2. No config changes option A LudwigConfig(input_features=[ BinaryInputDenseEncoder(name="foo", num_layers=3, preprocessing=...) ]) # 3. No config changes option B...

Current thoughts: - Generally like the nested structure - Option 2 may be the best compromise for now - Worth investigation to see how hard option 1 would be, as...

@brightsparc we discussed the challenge of implementing the nested structure in terms of the refactoring effort. The consensus is that we should explore it and see where the complexity exists,...

Thanks for raising this issue @Jeffwan. This should be relatively quick to implement, so we'll see if we can get it added for v0.6.