The Banana effect

Not true, researching and training frontier models cost millions if not billions. If Meta did not release the LLAMA weights we would have had a chance to regulate.
Open sourcing small narrow models is overall a very good idea, open sourcing frontier models was not.

Now I'll never get those nice medals..

Hard disagree. I really don’t understand how people think open source frontier models should not be regulated. The world has gone completely mad.

Same same

That was my New Year’s Eve

Favour either becoming a theory god, a builder of useful stuff, or an expert experimenter who designs solid interventions. If you really must, a data-crunching modeler, but with a relentless focus on predictive modelling.

It may also be because the project you’re trying to install causal-pipe into has other constraints for the dependencies, newer or older versions of numpy for example. Then it could be a bit complex to find the right version that works for all dependencies, you need to look at the pip error trace.

Depending on your machine, some pip packages may need to be compiled from source. That may be what you’re seeing.
What is the full pip error? What’s your machine/operating system?
You also need R if you want to use all the capabilities of the library.

Wrong link indeed ! ;) thank you for pointing this out

pypi.org/project/caus...

Github: github.com/albertbuchar...

Docs/PyPI: pypi.org/project/causal…#CausalDiscoveryr#CausalInferencece

- Cyclic SCM solver for simulation/fit checks
- Model selection via pseudolikelihood + MMD^2

Causal discovery has loads of trade-offs, but the workflow feels powerful.

Next up: faster PySR hillclimb, longitudinal data, MSM with automated nuisance/weight models for IPTW, IPCW, AIPW.

- PySR integration (amazing library by @MilesCranmer) with chainable hill climbing: SEM → best graph → PySR, or run PySR directly.q
- FAS bootstrap edge inclusion probabilities for edge stability quantification and filtering
- With FAS/KCI → PySR works nicely for nonlinearities