Getting Started

Hello-world

print "Hello-world"

Comments

Comments start with a #, and DSL-FR will ignore them:

# this is a comment
print "Hello-world"

Types

Basic types

Array types

Arrays are containers that hold values all of the same type.

a1 = [7,8,9]

The indexes of the array are all integers, from the front to the back to increase from 0, like 0, 1, 2, 3… From the back to the front, they are negative, and decreases from -1, like -1, -2, -3 …

v1 = a1[ 2 ]  # v1 will be 9
v2 = a1[ -1]  # v2 will also be 9

Dictionary type

Dictionaries are containers that associate keys of a given type with values of another type. They are can grow dynamically.

dic1 = {
	"filename": "car.png",
	"date": "Aug 23, 2020"
}
print filename of dic1  # will output "car.png"

Set type

Sets are containers that only store the unique values they have seen. They are can grow dynamically.

s1 = set of [ 1, 2, 2, 3]  # s1 will be [1,2,3] because duplicate elements will be ignored

Domain specific types

• Image

• DetectedObject

Control Flow

• if-else

if var1 > var2 
    ...
else if var1 < var2 
    ...
else
    ...

Loops

• foreach

foreach image in images
    ...

• repeat

repeat 5 times
    ...

Quantifiers

• exists

As long as one image meets the condition, execute the following code.

exists img in images that img.contain(dog)
    ...

• ifall

Only all images meet the condition, execute the following code.

ifall img in images that img.contain(dog)
    ...

Functions

• function

Functions in DSL-FR are similar with Ruby(a general programming language). When calling a function, the parentheses can be omitted.

Preposition parameters

New

When define a function, the main parameters are defined in parentheses after the function name. The remaining parameters are passed in as prepositions.

The supported prepositions are in, of, from, to, with.

Multiple parameters are separated by commas.

For example, if a function is used to filter pictures within a specified time range, it can be defined as follows:

function filter( images )
    from : date1
    to : date2
    
    ... # filtering...

    return filtered_images

Then, you can call this function like this:

imgs = filter images from "Jan 1, 2020" to "Sept 1, 2020"

Domain specific functions

• get()

Get an attribute of an object.

function get( attr [, attr2...] ) of obj
    ...
    return attr

• select()

Detect and return objects from images.

function select( obj_name [, obj_name2...] ) from images
    ...
    return objects

• mark()

Identify objects from images and mark the boundaries of the objects.

function mark( obj_name [, obj_name2...] ) in images
    ...
    return images

Other built-in functions

• calculate()

Calculate some indicators of an object.

function calculate( indicator ) of obj
    ...
    return value

Modules

User-defined modules

A module is composed of classes and methods. A module is composed of classes and functions. Write them into a file, and the file name (without extension) is the module name.

Type extensions

DSL-FR provides an extend keyword that allows programmers to add and modify methods of various types at compile-time, including built-in types like int or string.

extend image
    function save( filepath )
        ... # save image to the given filepath

Aggregators

• minimum, maximum, mean

Aggregators are defined in the module “math”.

import math

print math.maximum of [1,2,3]  # will output 3

Visualizations

• draw

draw histogram with { 
    'x_axis': stats of persons
}

draw scatter_plot with { 
    'x_axis': stats of persons, 
    'y_axis': stats of smiles
}

Pipelines

New

Basic Pipelne

For some research questions, we can subdivide each research question into small steps to solve it like a pipeline. In order to reduce intermediate calculations, we use keyword pipeline and symbol |> to abstract this process. A basic pipeline is as follows:

pipeline
    open "/paht/to/images" |> select smiles, persons |>
    draw scatter_plot with { 
        title:  "The relationship between the persons and smiles.",
        x_axis: persons,
        y_axis: smiles
    }

Advanced Pipeline

For some complex research questions, we have a variety of tasks to deal with, but they are not related to each other, they can be processed in parallel. For such a pipeline containing multiple parallelizable processing tasks, we call it Advanced Pipelines.

We use keyword stage to tag each task with a name. The result of each stage will be temporarily saved and can be used as the input of other tasks. For example:

pipeline
    stage s1
        open "/paht/to/images1" |> select smiles

    stage s2
        open "/paht/to/images2" |> select persons
    
    s1, s2 |>
        draw scatter_plot with { 
            title:  "The relationship between the persons and smiles.",
            x_axis: persons,
            y_axis: smiles
        }

File operations

• open

Using built-in function open, we can open an image or the images in a folder by gicen a filepath.

image = open "car.png" 
    ... # do something with image

image = open "path/to/images/" 
    ... # do something with image

Training a new model

TODO

Debugging

TODO