Skip to content

Latest commit

 

History

History
244 lines (185 loc) · 7.08 KB

README.md

File metadata and controls

244 lines (185 loc) · 7.08 KB

spacekit

GitHub license CodeFactor Build Status Powered by STScI Badge Powered by Astropy DOI

Astronomical Data Science and Machine Learning Toolkit

ML Dashboard

Setup

Install with pip

# install extra deps for all non-pipeline tools (analysis, training, data viz)
$ pip install spacekit[x]

# for bare-minimum dependencies (STScI/SDP pipeline operations):
$ pip install spacekit

Install from source

$ git clone https://github.com/spacetelescope/spacekit
$ cd spacekit
$ pip install -e .[x]

Testing

See tox.ini for a list of test suite markers.

# run all tests
$ pytest

# specify the `env` option to limit tests to a specific 'skope'
# env options: "svm", "hstcal", "jwstcal"
$ pytest --env svm -m svm
$ pytest --env hstcal -m cal
$ pytest --env jwstcal -m jwst

Pre-Trained Neural Nets

JWST Calibration Pipeline Resource Prediction (JWST)

JWST CAL Docs

  • Inference spacekit.skopes.jwst.cal.predict

From the command line:

$ python -m spacekit.skopes.jwst.cal.predict /path/to/inputs

# optionally specify a Program ID
$ python -m spacekit.skopes.jwst.cal.predict /path/to/inputs --pid 1076

From python:

> from spacekit.skopes.jwst.cal.predict import JwstCalPredict
> input_path = "/path/to/level1/exposures"
# optionally specify a Program ID `pid` (default is None)
> jcal = JwstCalPredict(input_path, pid=1076)
> jcal.run_inference()
# estimations for L3 product memory footprints (GB) are stored in a dict under the `predictions` attribute. Ground truth values (latest actual footprints recorded) are shown as inline comments.
> jcal.predictions
{
    'jw01076-o101-t1_nircam_clear-f212n': {'gbSize': 10.02}, # actual: 10.553384 
    'jw01076-o101-t1_nircam_clear-f210m': {'gbSize': 8.72},  # actual: 11.196752
    'jw01076-o101-t1_nircam_clear-f356w': {'gbSize': 7.38}, # actual: 6.905737
}
# NOTE: the target number "t1" is not intended to match actual target IDs used by the pipeline.

Single Visit Mosaic Alignment (HST)

SVM Docs

  • Preprocessing: spacekit.skopes.hst.svm.prep
  • Predict Image Alignments: spacekit.skopes.hst.svm.predict
  • Train Ensemble Classifier: spacekit.skopes.hst.svm.train
  • Generate synthetic misalignments†: spacekit.skopes.hst.svm.corrupt

† requires Drizzlepac

HST Calibration Pipeline Resource Prediction (HST)

HST CAL Docs

  • Training spacekit.skopes.hst.cal.train
  • Inference spacekit.skopes.hst.cal.predict

Exoplanet Detection with time-series photometry (K2, TESS)

K2 Docs

  • spacekit.skopes.kepler.light_curves

Customizable Model Building Classes

Build, train and experiment with multiple model iterations using the builder.architect.Builder classes

Example: Build and train an MLP and 3D CNN ensemble network

  • continuous/encoded data for the multi-layer perceptron
  • 3 RGB image "frames" per image input for the CNN
  • Stack mixed inputs and use the outputs of MLP and CNN as inputs for the final ensemble model
ens = BuilderEnsemble(XTR, YTR, XTS, YTS, name="svm_ensemble")
ens.build()
ens.batch_fit()

# Save Training Metrics
outputs = f"data/{date_timestamp}"
com = ComputeBinary(builder=ens, res_path=f"{outputs}/results/test")
com.calculate_results()

Load and plot metrics to evaluate and compare model performance

Analyze and compare results across iterations from metrics saved using analyze.compute.Computer class objects. Almost all plots are made using plotly and are dynamic/interactive.

# Load data and metrics
from spacekit.analyzer.scan import MegaScanner
res = MegaScanner(perimeter="data/2022-*-*-*")
res._scan_results()

ROC

Eval

Preprocessing and Analysis Tools for Space Telescope Instrument Data

box

from spacekit.analyzer.explore import HstCalPlots
res.load_dataframe()
hst = HstCalPlots(res.df, group="instr")
hst.scatter

scatter

spacekit
└── spacekit
    └── analyzer
        └── compute.py
        └── explore.py
        └── scan.py
        └── track.py
    └── builder
        └── architect.py
        └── blueprints.py
        └── trained_networks
    └── dashboard
        └── cal
        └── svm
    └── datasets
        └── _base.py
        └── beam.py
        └── meta.py
    └── extractor
        └── load.py
        └── radio.py
        └── scrape.py
    └── generator
        └── augment.py
        └── draw.py
    └── logger
        └── log.py
    └── preprocessor
        └── encode.py
        └── ingest.py
        └── prep.py
        └── scrub.py
        └── transform.py
    └── skopes
        └── hst
            └── cal
                └── config.py
                └── predict.py
                └── train.py
                └── validate.py
            └── svm
                └── corrupt.py
                └── predict.py
                └── prep.py
                └── train.py
        └── jwst
            └── cal
                └── config.py
                └── predict.py
        └── kepler
            └── light_curves.py
        
└── pyproject.toml
└── setup.cfg
└── tox.ini
└── tests
└── docker
└── docs
└── scripts
└── LICENSE
└── README.md
└── CONTRIBUTING.md
└── CODE_OF_CONDUCT.md
└── MANIFEST.in
└── bandit.yml
└── readthedocs.yaml
└── conftest.py
└── CHANGES.rst
                       
           /\    _       _                           _                      *  
/\_/\_____/  \__| |_____| |_________________________| |___________________*___
[===]    / /\ \ | |  _  |  _  | _  \/ __/ -__|  \| \_  _/ _  \ \_/ | * _/| | |
 \./    /_/  \_\|_|  ___|_| |_|__/\_\ \ \____|_|\__| \__/__/\_\___/|_|\_\|_|_|
                  | /             |___/        
                  |/