HackerTrans
TopNewTrendsCommentsPastAskShowJobs

madman2890

no profile record

Submissions

Relational signal compression across your data lake

kurve.ai
1 points·by madman2890·4 mesi fa·0 comments

Why Most Valuable AI Systems Are Still Tabular Models

6 points·by madman2890·4 mesi fa·2 comments

Predictive AI with graphs and tabular foundation models

wesmadrigal.github.io
2 points·by madman2890·4 mesi fa·0 comments

Tabular data is the frontier – graphs can help

5 points·by madman2890·4 mesi fa·2 comments

Tabular Foundation Models Still Need One Thing: Multi-Table Aggregation

wesmadrigal.github.io
2 points·by madman2890·4 mesi fa·1 comments

[untitled]

1 points·by madman2890·2 anni fa·0 comments

[untitled]

1 points·by madman2890·2 anni fa·0 comments

Library for feature engineering across many table using graphs

wesmadrigal.github.io
1 points·by madman2890·2 anni fa·0 comments

[untitled]

1 points·by madman2890·2 anni fa·0 comments

[untitled]

1 points·by madman2890·2 anni fa·0 comments

[untitled]

1 points·by madman2890·2 anni fa·0 comments

[untitled]

1 points·by madman2890·3 anni fa·0 comments

[untitled]

1 points·by madman2890·3 anni fa·0 comments

Feature Engineering Abstractions with Graphs

github.com
2 points·by madman2890·3 anni fa·0 comments

[untitled]

1 points·by madman2890·3 anni fa·0 comments

[untitled]

1 points·by madman2890·3 anni fa·0 comments

[untitled]

1 points·by madman2890·3 anni fa·0 comments

[untitled]

1 points·by madman2890·3 anni fa·0 comments

Show HN: built a feature engineering interface based on graphs

github.com
1 points·by madman2890·3 anni fa·1 comments

comments

madman2890
·4 mesi fa·discuss
with all of it? wow :)
madman2890
·4 mesi fa·discuss
We need the government to release the alien technology to the public now.
madman2890
·4 mesi fa·discuss
Glad you find it useful.
madman2890
·4 mesi fa·discuss
Tabular foundation models like TabPFN and related work are extremely promising. They’re starting to show strong results on many classical tabular ML benchmarks and can reduce the amount of manual modeling work required from data scientists. However, there is a structural reality of enterprise data that these models don’t remove. Most real-world machine learning problems are not stored in a single clean table. Instead they live across dozens or hundreds of relational tables: orders, customers, events, transactions, shipments, products, logs, etc. Each table captures part of the signal, often with one-to-many relationships, time dependencies, and high cardinality entities. Before any tabular model can be trained, those signals have to be integrated. In practice this means: Traversing relational graphs of tables Aggregating child tables to parent entities Handling time windows and temporal leakage Collapsing many-to-many relationships into meaningful features Producing a single wide training dataset This step is usually the most time-consuming part of the entire ML workflow. Even if the model itself becomes automated via a tabular foundation model, the data still has to be prepared. This is where GraphReduce comes in. GraphReduce treats the relational database as a graph of entities and relationships. Instead of manually writing large SQL pipelines, the user defines the nodes (tables) and their relationships. GraphReduce then walks the graph and performs the required aggregations automatically, generating a single training dataset.
madman2890
·11 mesi fa·discuss
“Smart developer’s quirks” tend to peak in 3-8 years of experience and fade off thereafter. A hipster will never fade off and instead continue hipster coding alongside their identity in perpetuity.
madman2890
·2 anni fa·discuss
Way overdue, but finally got the docs up for my project graphreduce. Feedback, contributions, and stars welcome!
madman2890
·2 anni fa·discuss
In data pipelines or MLOps projects I've found that mirroring the code location in the artifact storage location, such as in object stores, to be a helpful pattern. This little lightweight library makes using this pattern a drop in for any project.
madman2890
·3 anni fa·discuss
Taking numerous raw tables to a flat ML/AI ready feature vector can be a challenge. This is a lighter weight way of doing it than feature stores, if it may serve you!
madman2890
·3 anni fa·discuss
I've found feature stores to be overkill for many projects, as I just need point-in-time correctness, abstractions for joining lots of tables and flattening to a specific granularity, and abstractions for composability. I've been building GraphReduce which helps with all the aforementioned. Hope it serves others!
madman2890
·3 anni fa·discuss
An early release of the tech - an automation layer for discovering inclusion dependencies, building a graph of data where tables/files are nodes and foreign keys are edges between them.
madman2890
·3 anni fa·discuss
working on a composable feature engineering abstraction for handling big pieces of FE pipelines: time travel, entity linkage, standardized interface, pluggable compute backends, and simplification of features with many tables
madman2890
·3 anni fa·discuss
I've been working on a composable feature engineering abstraction with graphs. Feature stores are too heavy for most projects I work on, but are complementary nonetheless.
madman2890
·3 anni fa·discuss
TLDR; graph data structures for feature engineering and data transformations - spark, dask, and pandas supported

Best way to get started while docs are still lacking: https://github.com/wesmadrigal/GraphReduce/blob/master/examp...
madman2890
·3 anni fa·discuss
The gym
madman2890
·3 anni fa·discuss
Alaska will soon be the most desirable state to live in.
madman2890
·3 anni fa·discuss
Actually, the risk is stratified by age. With this entire pandemic, risk has been stratified by age, co-morbidities, etc.

If you're a male under 40 years old your risk of myocarditis is higher from the vaccines. Not to mention, if you don't have co-morbidities, your risk of death from COVID is quite low. It's a simple cost benefit analysis, really. According to a linked study (ahajournals.org link), risk of cardiac related issues from vaccine for males under 40, like myself, is 1/10000 or 0.0001. Risk of cardiac related issues from covid-19 for males under 40 is about 0.0000016, so about a factor of 6 lower risk. Additionally, as we know from more recent studies, especially with the Kraken XBB variant, vaccines offer virtually no protection and almost everyone will get infected. So we can code probability of infection post-vaccination and post-natural infection as equal for the sake of simplicity. Additionally, there is a growing body of research discussing reinfection issues during a subsequent infection by vaccination status, and the latest numbers I extracted from the paper are coded below. In summary, for me, a 32 year old male, we get the following risks profiles:

vaccine risk = P(problem | vaccine, male under 40, no comorbidities) * P(covid | vaccine immunity) * P(problem second infection | vaccine immunity) vaccine risk = 0.0001 * 0.95 * 0.51 vaccine risk = 0.000004845

natural immunity risk = P(problem | covid-19, male under 40, no comorbidities) * P(covid | natural immunity) * P(problem second infection | natural immunity) natural immunity risk = 0.0000016 * 0.95 * 0.47 natural immunity risk = 0.00000007144

vaccine risk vs. natural risk

Vaccine 67X more risky than natural route (FOR ME)

Myocarditis risk vaccines vs. natural immunity males under 40: https://www.ahajournals.org/doi/abs/10.1161/CIRCULATIONAHA.1...

XBB 1.5 Kraken likelihood of infection: https://pubmed.ncbi.nlm.nih.gov/36580913/

Serious issues associated with subsequent infection by vaccination status: https://www.thelancet.com/journals/lanmic/article/PIIS2666-5...
madman2890
·3 anni fa·discuss
This part means that natural immunity outperforms vaccine immunity in protecting against reinfection and spread of COVID-19 over time.

" Discussion

Our study shows that the mRNA-containing LNP vaccines BNT162b2 and mRNA-1273 induce high anti-S1 IgG and IgA levels in the blood as well as in the saliva, but these Ig levels steadily decrease over time and approach levels that are comparable to the long-term levels induced by two immunizations with the adenovirus-based vaccine AZD1222. In the long run, such pronounced anti-S1 IgG and (s)IgA reductions in the saliva likely reflect the declining protection against infection and from spreading in the respiratory tract of naïve individuals (16, 17). On the other hand, the observed stronger anti-S1 (s)IgA response in the saliva of previously infected vaccinees – likely generated by re-activation of infection-induced local (s)IgA+ memory B cells – might explain their recently described higher protection from infection and spreading "

These sections indicate that natural immunity is more effective at preventing infection and spread of COVID-19 than vaccines.

"In summary, the data indicate that the high initial mRNA vaccine-induced anti-S1 IgG(1) and IgA responses decrease over time and approach levels induced with the adenovirus-based vaccine up to day 270. Higher and more stable anti-S1 (s)IgA levels in the saliva of pre-infected vaccinees might explain their higher protection from infection and spread of SARS-CoV-2.

Intriguingly, the mRNA vaccines, and in particular the mRNA-1273 vaccine, induced increasing long-term anti-S1 serum IgG4 levels in naïve individuals with hitherto unclear influences on the fight against the pathogen. Naïve individuals vaccinated with the adenovirus-based vaccine did not show such long-term anti-S1 IgG4 response at least after two vaccinations until day 270. "
madman2890
·4 anni fa·discuss
Where is the farm at? I own one in Manizales, Colombia. We have around 25,000 trees.
madman2890
·4 anni fa·discuss
Manizales, Colombia
madman2890
·5 anni fa·discuss
I bought a coffee farm in Colombia. Annualized it makes around $800 / month and growing. Should double to $1600 / month next year as we’ve doubled our production. It’s barely profitable, but is profitable nonetheless. The more we plant the more profitable it will get, and we’ll be planting 10,000 more trees soon!