Pip installation outdated?

General
1

Hello,
I was trying to install BayesFlow from pip but it seems to be outdated regarding to the example notebooks.
For example, in the Two Moons notebook, one create a simulator like this

simulator = bayesflow.make_simulator([theta_prior, forward_model])

but running

bayesflow.__dir__()

output the following :

['__name__', '__doc__', '__package__', '__loader__', '__spec__', '__path__', '__file__', '__cached__', '__builtins__', 'logging', 'default_settings', 'exceptions', 'helper_functions', 'computational_utilities', 'losses', 'wrappers', 'helper_networks', 'coupling_networks', 'inference_networks', 'attention', 'summary_networks', 'networks', 'amortizers', 'diagnostics', 'configuration', 'helper_classes', 'simulation', 'trainers', 'sensitivity']

So no make_simulator function and it doesn’t look like the git repo architecture.

Is pip installation outdated ? Should I always work from source ?

My python version is Python 3.11.11 and I tried with Python 3.12.3 as well.

2

For now you would need to install from source before we make the official v2 “release”. For instance, to install from dev, you can do:

pip install git+https://github.com/bayesflow-org/bayesflow.git@dev

1 Like
3

Thanks ! Is dev is the preferrential working branch ? What is the branch policy for the project ? I didn’t find it.

4

The usual - main always has the most stable version of the project; dev may contain new features that are not yet in main, but perhaps not thoroughly tested.