What is a resource ?

A Resource is a fundamental object used in any Brick.

It represents data used or generated by tasks, or manually uploaded by a user in the DataBox. A task only takes resources as inputs and generates resources as outputs. A resource contains different Views that are used to generate visualizations of the resource (like chart or table).

For example, the object Table is a resource that stores 2d array data with tags and can be viewed as Excel-like sheet, Line plots, Bar plots, etc. 

How to create a resource ?

To create a new resource in your brick, you need to create a new class that:

• Extends the Resource class (import from gws_core)
• This class contains few methods to simplify resource management.
• Is decorated with the @resource_decorator (import from gws_core)
• To see how to define the name, description and style of the resource with the decorator view this page : Decorator
• Does not have mandatory parameters in the constructor and the constructor must call the parent constructor. This is because the system instantiates resources without parameters.

Your resource class must meet these conditions, otherwise the resource will be ignored by the system. 

Here is an example of our Robot resource.

from gws_core import resource_decorator
from gws_core import Resource
@resource_decorator(unique_name="Robot", human_name="Robot")
class Robot(Resource):
    age: int = IntRField()
    position: List[float] = ListRField()
    weight: float = FloatRField()

RField

All output resources of a task are automatically saved after the task is run. The system will save only the attributes of the class that are marked as an RField and will ignore all other attributes.

Here is an example of a string attribute marked as an RField that will be saved in the Resource. 

my_attribute = StrRField(default_value="Hello world")

The my_attribute will be saved after the task has been run and the system will automatically provide it when the resource is used in another task or is viewed.

There are 2 types of RFields :

• Searchable: this is only for small data. It will be stored in the database and therefore it can be searched using a full text search in the Databox
• Non searchable: this is for medium and large data. It will be stored in a file and the non-searchable files are lazy loaded (only loaded when there are requested) when the resource is reused.

Here is the list of the available RField to save different types of data. 

Primitive RFields

• IntRField : RField to save int values, by default they are searchable
• FloatRField: RField to save float values, by default they are searchable
• BoolRRifled : RField to save bool values, by default they are searchable
• StringRField : RField to save str values, by default they are not searchable

Object RFields

• ListRField : RField to save json list. It does not support python classes, please only provide serializable objects.
• DictRField : RField to save json dictionary. It does not support python classes, please only provide serializable objects.
• SerializableRField : RField to save python object. The python class of the object must be passed to the SerializableRField. The python class must extend the class SerializableObjectJson , have an empty constructor and implements the methods :
• serialize : method called when saving the resource to convert the python object to json to be stored.
• deserialize : class method called when the resource object is created and the RField loaded to instantiate the python object from the previously generated json object.
• Special RFields FileRfield : RField to dump and load resource directly to and from a file . You must create a class that extends the FileRField to use it (you can check the code of the DataFrameRField that uses the FileRField to dump DataFrame into a file using read_pickle function). The sub class of FileRfield must implement 2 methods :
• dump_to_file : method called when saving a resource to save the RField to a file. The RField and file path are passed to this method (dump the resource in the provided file, not elsewhere).
• load_from_file : method called when the resource object is created and the RField loaded to instantiate the Rfield from the file.
• DataFrameRField : RField to save pandas Dataframe object in a file. WARNING : It uses the “read_pickle” function to deserialize the Dataframe, so only use it for one Dataframe that you trust.
• ResourceRField : RField to link another resource inside a resource. The associated resource MUST be an input of the task that generated the resource containing the ResourceRField.

Name a resource

Each resource object has a "name" property. This property allows you to set a custom name for your created resource in your task. This name will be visible both in the databox and in the experiment playground. Users can always update this name from the interface.

To define the name, simply set the "name" attribute of your resource object:

my_resource.name = 'New custom name'

Technical information

You can add technical information to a resource to attach more information to the resource. These informations are always attached to the views and are visible when a user generates a view.

The technical information object is a dictionary of TechnicalInfo objects. A TechnicalInfo is a key/value object. More attributes will be added to it later.

To attach a TechnicalInfo to a resource, create the TechnicalInfo object :

technical_info = TechnicalInfo(key='key', value='value')
And then add it to the resource by calling : 
my_resource.add_technical_info(technical_info)

The technical information will be automatically saved and accessible on the resource's views. You can also get the technical information by calling get_technical_info(key: str): TechnicalInfo

Views

A resource can have different views, to allow the user to navigate through the data of the resource with different views.

See Resource Views section for more details.