The most annoying thing about software engineering
Is how you spend 2 weeks learning intensely about some topic (e.g. how to do graceful shutdown properly),
And 6 months later you completely forget everything 😡
@chris_mccord Just tested on https://t.co/0nPEK4PuFa:
~3x faster on average
sometimes from 3.6ms down to 0.5ms
This is on a very frequently updated LiveView (renders cursors on every mouse move)
This is amazing.
What if you could:
* upload a .csv file to see & plot its contents
* stream and transform GBs of .parquet files from S3
* drag and drop a sqlite3 database and query it
* drag and drop an audio/video file for live transcription
With @livebookdev, you can: https://t.co/xMIzQcJRJd
Both Kyle and I (Trevor) make our TikToks and I finally hit a milestone of 1000 on one of my posts! Come follow us for some music goofiness!
https://t.co/rTmGCH8Uau
This is another huge win from HEEx templates: since we know the HTML structure and we know when it has changed (or not), we can skip huge parts of DOM parsing and traversal when whole subtrees stay the same.
@aloukissas@elixirlang It easy to build Elixir API clients with Tesla: https://t.co/KkG5F7sBez
And you don’t have to implement every function in the service’s API, just the things you use.
This is on @flydotio NVIDIA A100-SXM4-80GB .
Soon this kind of thing will be a `fly deploy` away from some GPU 🔥 . I'm having a blast playing with this stuff!
https://t.co/CDOWHbsHrd
This is going to change databases!
There are 1.7 million deployments of PostgreSQL worldwide. This makes it one of the world's most popular relational database management systems.
The team behind @postgresml reached out to me and showed me their open-source extension, PostgresML.
It's pretty cool!
You can use this extension to train a model on text and tabular data using SQL queries.
This example trains a model using XGBoost using a query!
select * from pgml.train('Handwritten Digit Image Classifier',
algorithm => 'xgboost',
'classification',
'pgml.digits',
'target'
);
The extension integrates Hugging Face 🤗 Transformers. You can use it to run NLP models right into the database.
Here is an example to determine the sentiment of some text:
select pgml.transform(
task => 'text-classification',
inputs => ARRAY[
'This product didn't work for our family.',
'This has been the best purchase of the year!'
]
) as positivity;
I had never seen this before.
You can train and run supervised and unsupervised algorithms. You can run pre-trained models. Machine Learning and AI right at the database level!
Here is the GitHub repository: https://t.co/WOmAy6DE55
This post is sponsored by @postgresml and their fully managed cloud service. You can use them in addition to their open-source library. Horizontal scalability and support for JavaScript, Python, and Rust.
Machine learning is now getting everywhere.
How long before this is part of every database management system out there?
@ZacharyNDean IMO Option 1 is the most future proof in terms of refactors. However, can still go w/ your preference and always use tools like recode/sourcer to normalize pre refactor later
Phoenix LiveView 0.20.0 is out with some of the features I showed off at ElixirConf:
🔥 assign_async / start_async / <.async_result>
🔥 HTML debug comments in your markup tree to know which component rendered any given fragment