Citrine

**What’s New**
* Training data are specified using a list of data sources, instead of a single data source. If multiple training data sources are specified, the data from all sources in the list will be used to train the predictor.
* Training data can be specified on a graph predictor. Any data sources specified by the graph predictor will be shared by all predictors in the graph, and shared data sources do not need to be redefined by any sub-predictors that require these data.
* You can now transfer the ownership of a project resource using `project.transfer_resource`.

**Deprecated**
* Specifying training data as a single data source. Please use a list of data sources instead, e.g. update`training_data=data_source` to `training_data=[data_source]`.

This is a small release focused on improving documentation and making the behavior of the Citrine Platform more transparent and explicit.

What's New
* The association between a GEM Table Data Source and the Formulation Descriptor that it provides is now explicit (https://github.com/CitrineInformatics/citrine-python/pull/422)

Improvements
* Improved documentation of when cross validation can be applied (https://github.com/CitrineInformatics/citrine-python/pull/427)
* Improved documentation of how to fetch the output descriptors of simple mixture predictors (https://github.com/CitrineInformatics/citrine-python/pull/347)

What's New
* A core element of machine learning is making sure inputs are all tabular -- sets of homogeneous vectors. GEMD represents real world data with graphs that can vary a lot, so it's essential to be able to bring consistency to material histories to train models. We are proud to announce the official naming of our graph-to-vector technology, GEM Tables, renamed from the working name, Ara.

Improvements
* The version number of a workflow execution is now visible. See [the documentation](https://citrineinformatics.github.io/citrine-python/reference/citrine.resources.workflow_executions.html) for details.
* Error messages, such as trying to train on an unregistered resource, have been improved.

Fixes
* Some objects were not receiving `id`s before being passed to the server. That's been fixed.
* Control over the scope in some variable definitions hadn't been included; that is now available.
* And some additional bug fixes, improvements to code structure and corrections to documentation.

What's New
* This release is largely about exposing [gemd-python v0.12.0](https://github.com/CitrineInformatics/gemd-python/releases/tag/v0.12.0) and modifying the citrine-python method signatures to mimic the changes therein.
* We are migrating Object constructors to require keyword arguments for nearly all parameters (excluding name), as per [our documented codestyle](https://github.com/CitrineInformatics/citrine-python/blob/master/CONTRIBUTING.md#codestyle).
* The name and labels arguments to the IngredientRun constructor have been deprecated for quite some time, and so those arguments have been removed.

Fixes
* Some errors in documentation were corrected along the way.

Note this release includes only usability improvements of alpha features.

Improvements
* Added `build_from_config` method to the tables collection, which allows you to get build and wait for the table to build in one method call. You can also use `initiate_build` and `get_by_build_job` for more advanced workflows.

Deprecated
* The `build_ara_table` and `get_job_status` methods of the table configs collection (aka Ara definitions) are now deprecated in favor of these new methods.

What’s New
* We've introduced a new API for the `ExpressionPredictor`. This new API allows bounds to be enforced for expression inputs and removes the need for duplicate descriptors in the graph. To use the new format, add a key to `aliases` for each unknown expression argument and map to its associated descriptor. For more details on how to migrate see the [docs](https://citrineinformatics.github.io/citrine-python/reference/citrine.informatics.predictors.html#citrine.informatics.predictors.DeprecatedExpressionPredictor).

Deprecated
* The previous ExpressionPredictor has been deprecated and renamed DeprecatedExpressionPredictor. Moving forwards, ExpressionPredictor will require aliases for every argument in the expression, and aliases will point to descriptors (previously, aliases pointed to descriptor keys). To migrate to the new ExpressionPredictor, create a new aliases dictionary mapping every argument in your expression to a descriptor (instead of a descriptor key). To continue using the old format, replace all instances of ExpressionPredictor with DeprecatedExpressionPredictor.