Data types and shapes
Each "line" that you connect in your network, that is, each connection between one port on one node to one port on another node, represents a value (also called a "tensor") that will be calculated once you evaluate or train your model. Each one of these "lines" has one data type and one shape. The data type and shape give us information about what the data will look like once we actually run the model. The data type and shape is displayed at each input and output port on a node.
The data type and shape that is being output by a node depends on the type of node and its configuration options. For some nodes, like inputs, constants and Rand, you can select the data type to use by clicking the node and configuring it on the right panel. Some other nodes, such as Sum and MatrixMult, will output the same data type as was provided as the input to the model. Check out the documentation of each node to see how they behave.
Data type
The available data types are:
- Int8
An 8-bit signed integer, capable of storing values from -128 to 127.
- Int16
A 16-bit signed integer, capable of storing values from -32768 to 32767.
- Int32
A 32-bit signed integer, capable of storing values from -2147483648 to 2147483647.
- Int64
A 64-bit signed integer, capable of storing values from -9223372036854775808 to 9223372036854775807.
- Uint8
An 8-bit unsigned integer, capable of storing values from 0 to 255.
- Float16
A 16-bit floating point number.
- Float32
A 32-bit floating point number.
- Float64
A 32-bit floating point number.
- Bool
A value which is either true or false.
- String
Text, in UTF-8 encoding.
Shape
The shape of a tensor describes its size. A tensor can be seen as a multi-dimensional array, and the shape is a list of numbers, for example: "(1, 2, 3)", which encodes the amount of elements in each dimension of a tensor. For example, the empty shape, "()", means that the tensor is just a single element, for example a single Uint8. The shape "(2)" means that the tensor is a list which contains two values, for example two strings. The shape "(3, 2)" means that the tensor contains three lists that each contain two elements, for a total of six elements. Sometimes a shape can vary - for example if you have images of different sizes. In this case, the shape is written with a question mark in the unknown dimension. For example, for an image with three channels but unknown width and height, the shape would be written "(3, ?, ?)".
