deepskies/DeepBench

Simulation library for very simple simulations to benchmark machine learning algorithms

What is it?

Simulation library for very simple simulations to benchmark machine learning algorithms.

Why do we need it? Why is it useful?

1. There are very universally recognized scientifically meaningful benchmark data sets, or methods with which to generate them.
2. A very simple data set will have objects, patterns, and signals that are intuitively quantifiable and will be fast to generate.
3. A very simple data set will be a great testing ground for new networks and for newcomers to practice with the technology.

Documentation

ReadTheDocs

To build from source

pip install sphinx
cd docs
make html

The folder docs/_build/html will be populated with the documentation. Navigate to file:///<Path To DeepBench>/docs/_build/html/index.html in any web browser to view.

Requirements

• python = ">=3.8,<3.11,"
• numpy = "^1.24.3"
• matplotlib = "^3.7.1"
• scikit-image = "^0.20.0"
• astropy = "^5.2.2"
• autograd = "^1.5"
• pyyaml = "^6.0"

Install

From PyPi

pip install deepbench

From Source

git clone https://github.com/deepskies/DeepBench.git
pip install poetry
poetry shell
poetry install
poetry run pytest --cov

General Features

1. very fast to generate
2. Mimics in a very basic / toy way what is in astro images
3. Be fully controllable parametrically

Included Simulations

1. Astronomy Objects - simple astronomical object simulation

• Galaxy, Spiral Galaxy, Star

2. Shapes - simple 2D geometric shapes

• Rectangle, Regular Polygon, Arc, Line, Ellipse

3. Physics Objects - simple physics simulations

• Newtonian Pendulum, Hamiltonian Pendulum

Example

Standalone

• Produce 3 instance of a pendulum over 10 different times with some level of noise.

import numpy as np
from deepbench.collection import Collection

configuration = {
	"object_type": "physics",
	"object_name": "Pendulum",
	"total_runs": 3,
	"parameter_noise": 0.2,
	"image_parameters": {
		"pendulum_arm_length": 2,
		"starting_angle_radians": 0.25,
		"acceleration_due_to_gravity": 9.8,
		"noise_std_percent":{
			"acceleration_due_to_gravity": 0
        }
    },
    "object_parameters":{
        "time": np.linspace(0, 1, 10)
    }
}

phy_objects = Collection(configuration)

phy_objects()

objects = phy_objects.objects
parameters = phy_objects.object_params

• Produce a noisy shape image with a rectangle and an arc

import numpy as np
from deepbench.collection import Collection

configuration = {
	"object_type": "shape",
	"object_name": "ShapeImage",
	"total_runs": 1,
	"image_parameters": {
		"image_shape": (28, 28),
		"object_noise_level": 0.6
	},
	"object_parameters": {

        "rectangle":{
            "object": {
                "width": np.random.default_rng().integers(2, 28),
                "height": np.random.default_rng().integers(2, 28),
                "fill": True
            },
            "instance": {}
        },
        "arc":{
            "object": {
                "radius": np.random.default_rng().integers(2, 28),
                "theta1":np.random.default_rng().integers(0, 20),
                "theta2":np.random.default_rng().integers(21, 180)
            },
            "instance":{}
        }
    }
}

shape_image = Collection(configuration)
shape_image()

objects = shape_image.objects
parameters = shape_image.object_params

Fine-Grained Control

• Make a whole bunch of stars

from deepbench.astro_object import StarObject
import numpy as np

star = StarObject(
        image_dimensions = (28,28),
        noise_level = 0.3,
        radius= 0.8,
        amplitude = 1.0
    )

generated_stars = []
x_position, y_position = np.random.default_rng().uniform(low=1, high=27, size=(2, 50))
for x_pos, y_pos in zip(x_position, y_position):
	star_object = star.create_object(x_pos, y_pos)
	generated_stars.append(star_object)

Contributions

Original Team

1. Craig Brechmos
2. Renee Hlozek
3. Brian Nord

Refactor and Deployment

1. Ashia Livaudais
2. M. Voetberg

Pendulum Team

1. Becky Nevin
2. Omari Paul

Contributing

Please view the deepskies contribution guidelines before submitting a code addition

Acknowledgement

This work was produced by Fermi Research Alliance, LLC under Contract No. DE-AC02-07CH11359 with the U.S. Department of Energy. Publisher acknowledges the U.S. Government license to provide public access under the DOE Public Access Plan DOE Public Access Plan. Neither the United States nor the United States Department of Energy, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any data, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights.

We acknowledge the Deep Skies Lab as a community of multi-domain experts and collaborators who’ve facilitated an environment of open discussion, idea-generation, and collaboration. This community was important for the development of this project.