# Getting Started#

Halerium can be either used online on the Halerium platform or locally as a pure Python package. To get your test account for the Halerium platform visit https://erium.de/ For local usage follow the installation instructions:

## High level usage#

For the most common applications Halerium offers the convenient `CausalStructure`

class at the package level.
With this class you can easily create, train and evaluate Bayesian models with just a couple of lines.

```
from halerium import CausalStructure
causal_structure = CausalStructure([input_columns, output_columns])
causal_structure.train(training_data_frame)
causal_structure.predict(test_data_frame)
```

To learn more about the usage of the high-level package check out

## Low level usage#

With the halerium.core package you can build custom tailored Halerium graphs, with which you can model arbitrarily complex processes. You choose how much knowledge is hard-coded into your graph and how much is to be learned from data.

Build deep hierarchical graphs with Variables of arbitrary dimensions and complex tensor operations. Combine a graph with incomplete data and let halerium solve your model for you. Utilize the powerful combination of Bayesian inference and machine learning without caring about the statistical details!

```
import halerium.core as hal
with hal.Graph("g") as g:
hal.Variable("x", shape=(3, 5), mean=0., variance=1.)
hal.Variable("y", shape=(3,))
y.mean = hal.sigmoid(hal.sum(x, axis=1))
y.variance = hal.tensordot(x, x, axes=2)
model = hal.get_posterior_model(g, data={g.y: [1, 2, 3]})
model.get_means(g.x)
>>> array([[[0.14381036, 0.14378252, 0.14378723, 0.14380088, 0.14379767],
[0.32230996, 0.32230319, 0.3222988 , 0.3222934 , 0.32231929],
[0.41568274, 0.41570946, 0.41568607, 0.41570675, 0.41571199]]])
```

Check out the in-a-nutshell introduction in

Find more examples and the code overview in

The advantages of halerium.core compared to conventional machine learning are illustrated in the advantages section.