Carl Thomé

Is anyone working on this on Databricks' side? This is the primary feature holding me back before I can fully depend on MLflow.

This is sorely missing. I'm tuning both feature representation and model topology simultaneously in a keras-tuner experiment, and when the number of pooling layers is too many for the size...

Which is particularly confusing with the GCP deployer, as AI platform is still only on Python 3.5 AFAIK. This crashes with the same error: ```py import fairing GCP_PROJECT = fairing.cloud.gcp.guess_project_name()...

As a workaround, [this image](https://cloud.google.com/ai-platform/deep-learning-containers/docs/choosing-container) seems to work fine: `gcr.io/deeplearning-platform-release/tf-gpu.1-14`

ETA on this? I'm using a lot of numerical computing libraries that are really just FFIs for C libraries. They tend to be iffy to `pip install`and requirements.txt doesn't suffice.

I get the exact same error (via librosa.load as well) but it seems to happen randomly when loading several files in succession. It's hard to reproduce, but it happens frequently...

> @carlthome it could be an OutOfMemory. Yup, I think that was it actually. It would be nice if audioread gave clearer errors than NoBackendError.

Not really, no. I eventually started calling ffmpeg via `subprocess.Popen` and am pretty happy with more explicit control for my purposes. ```py import shlex import subprocess import numpy as np...

Status?

For many applications a little bleed might be acceptable, but in some any bleed at all might screw up the results. I'd prefer if the has_bleed boolean was clearly defined...