825 karma · joined August 11, 2017
The problem is that the shallow work now occupy most of your time, and since you are not doing deep work, you deep skills decay. The decay created a gap between what is expected from you, and what you are current level is.
Another issue is the great acceleration of new tools / new tech , which occur both in breath (a lot of new tools) and depth( each tool is more complex). This acceleration contribute to the imposter syndrome, since you do not invested enough time to become expert in the new tech. I.e. you were promoted based on you skills with the old tech.
What you must do, if you want to close the gap, is use your free time to do more deep work.
If you are state-full you should prefer VPA.
If you are state-full and need HA you might want to combine both.
Another case for VPA is data processing, e.g. pandas, which might be easier to scale to bigger datasets without going into out of memory data frames.
My 2 cents, before leaving, I would seek more professional advice about this issues.
What is needed (I think) is some sort of voice based IDE, where the programmer will be able to navigate the code in 2D. This is not an easy fit.
Most of the Accessibility API are for reading, but it should be easy to navigate and create code.
Note that big companies do have blind programmers, so you might be able to contact them and do user research.
The issue with retraining is that you need new labels (assume supervised ML). so I wonder what process do you use to get those.
So I assume that you are doing hyperparameter search? Can you share what optimization method you are using for search (e.g. random, gp )?
Also, is the search can be distributed in parallel to multi node ?
And, if mindsdb is not part of the db, what happen if minddb fail ?
Also, do you support automatic retraining? If yes, can you elaborate more?
Second you should not give any sort of control of the board (not the company).
And you will be fine.
Somebody (another hedge funds) found a flaw in the way RH is operating.
Also note that (I think) you cannot share GPU between Pods. So you might want to have different JH containers inside the same Pod (with different ports).
Are you looking for EXECUTION of the notebook on a cloud VM? I.e. as a k8s job?
Also, the rise of deep learning is based upon the rise of big labeled data.
Next, I would create microservices using GRPC and define an interface between them, this way they can continue working on each stack (and you can replace it in future date).
I assume that if you choose a unified stack there will be a rewrite which is risky.
Most of the new data systems (e.g. Snowflake) are build on top of a data lake (E.g. S3 bucket) and a scale out JIT compute nodes.
By using such tool, you can read the data from S3 once, and avoid loading the data into memory when you add nodes.
A specific use case that I am working on now is Auto ML. Imagine training 100's of models on the same data.
Also, startups have different stages. The networking/fundraising, etc. Actually comes AFTER product/market fit.