Delta Lake as source of images to train a classification model on a local computer

[MCosta]

Hi Folks,

I'm evaluating Delta Lake to store image / data version control to be used to train models. I looked at a session explaining how to do this and also using MLflow to manage training (https://databricks.com/session_na21/image-processing-on-delta-lake).

Note: it'd be interesting to have a link to the source code used in the demo.

I have a slightly different scenario, though. Testing is being performed on a local machine following the quick tutorial (https://docs.delta.io/latest/quick-start.html). In this scenario, what is the best way (using as much out-of-the-box components as possible) to "grab" a local folder with images organized into subfolders (classes) and dump them into delta lake and then use a specific snapshot on tensorflow?

Thanks

[Kaniz]

Hi @ MCosta! My name is Kaniz, and I'm the technical moderator here. Great to meet you, and thanks for your question! Let's see if your peers on the Forum have an answer to your questions first. Or else I will follow up shortly with a response.

[-werners-]

I can think of 3 ways for doing this:

1. using the web UI (the create table option or upload data into DBFS)
2. using databricks-connect, which bridges your local machine with the remote databricks clusters
3. using the databricks-cli to copy local files to dbfs

your cloud vendor might also have a tool to copy local data into the cloud environment.

For your purpose (evaluating) the web UI option might be the easiest.

https://docs.databricks.com/data/data.html

https://docs.microsoft.com/en-us/azure/databricks/data/databricks-file-system#file-upload-interface