README for classify.py / graph_clustering

Update history

September 2019: - HSC DR1 UDEEP catalogue added - Code w/ examples added

Near future: - Add DR2 DEEP/UDEEP catalogues with varying k - Add better documentation - Update the script to produce more user friendly catalogues

---

Reference: Martin 2019b

Contact: garrethmartin@arizona.edu

---

Purpose:

Clusters objects found in a list of astronomical images by their visual similarity. Objects are sorted into k groups and a catalogue containing object centroids, group number, size in pixels and silhouette score is output.

Prerequisites:

• numpy
• scipy
• astropy
• configobj
• sklearn
• joblib
• matplotlib
• dotnetcore SDK

Installation on Python 2.7 (3 not tested):

pip install graph_clustering

or build from source

python setup.py install graph_clustering

Usage:

Using the built-in script:

usage: classify.py [-h] [-bd BASE_DIR] [-dd DATA_DIR] [-im IMG_NAMES]
                   [-il IMG_LIST] [-ib BOUNDS] [-wl BANDS] [-oi OUT_ID]
                   [-pd PATCH_DIR] [-ps PATCH_SIZE] [-nn N_NODES]
                   [-ct HC_TARGET] [-nt N_THREADS] [-ns N_SAMPLES]
                   [-ni N_ITERATIONS] [-mt METRIC] [-pe] [-pl] [-ng] [-hc]
                   [-cc] [-gt]

optional arguments:
-h, --help            show this help message and exit
-bd BASE_DIR, --base_dir BASE_DIR
                      Base directory
-dd DATA_DIR, --data_dir DATA_DIR
                      Data directory
-im IMG_NAMES, --img_names IMG_NAMES
                      Image filenames
-il IMG_LIST, --img_list IMG_LIST
                      File containing list of image filenames
-ib BOUNDS, --bounds BOUNDS
                      Image bounds
-wl BANDS, --bands BANDS
                      Image bands
-oi OUT_ID, --out_id OUT_ID
                      Output id for the model
-pd PATCH_DIR, --patch_dir PATCH_DIR
                      Patch sub-directory
-ps PATCH_SIZE, --patch_size PATCH_SIZE
                      Patch radius in pixels
-nn N_NODES, --n_nodes N_NODES
                      Number of nodes in the GNG graph
-ct HC_TARGET, --HC_target HC_TARGET
                      Target clusters of HC clustering
-nt N_THREADS, --n_threads N_THREADS
                      Number of CPU threads
-ns N_SAMPLES, --n_samples N_SAMPLES
                      Number of samples to process before adding a new node
-ni N_ITERATIONS, --n_iterations N_ITERATIONS
                      Total number of GNG iterations
-mt METRIC, --metric METRIC
                      Metric used to measure distance [pspearson, cosine,
                      linear]
-pe, --extract        Perform extraction step
-pl, --parallel       Use with --extract to perform extraction step in
                      parallel
-ng, --gng            Perform growing neural gas step
-hc, --h_cluster      Perform heirarchical clustering step
-cc, --conn_comps     Perform connected component step
-gt, --galaxy_train   Perform k-means clustering on galaxy feature vectors
                      and produce catalogues
-mg, --montage        Create html montage of galaxy images
-kc K_CLUSTER, --k_cluster K_CLUSTER
                      Number of clusters to use for the montage step

By importing the package

from graph_clustering import run_classify
>>> run_classify(dir_base=dir_base, dir_data=dir_data, files=files, bounds=bounds,
         bands=bands, output_id=output_id, patch_folder=patch_folder,
         patch_size=patch_size, nodes=nodes, clustering_target=clusters,
         threads=threads, metric=metric, n_samples=n_samples,
         iterations=iterations,
         patch_extraction=True,
         parallel_extract=True,
         GNG_model=True,
         hierarchical_clustering=True,
         connected_components=True,
         group_train=True,
         make_montage=True,
         k=k)

See example folder for working example with HSC data