evevkovacs / ML-SN-Classifier

Machine-Learning Algorithm for Supernova Classification

ML-SN-Classifier

Machine-Learning Codes for Supenova Classification

Supernova Identification with Random Forest (SNIRF)

Code Overview

SNIRF.py performs photometric classification using a Random-Forest machine-learing algorithm. It builds and tests the classifier using user-supplied training and test data. The user can supply additional data files which will be classified by the code. The code saves the classifications in auxiliary text files.

The package consists of 3 modules:

• SNIRF.py (executable script to build and run the classifier and outputs results)
• ML_globals.py (contains definitions of global constants and functions)
• ML_plots.py (optional plotting code that makes a variety of plots)

Usage

python SNIRF.py [command-line options]

Command-Line Options

Many command-line options are available.
python SNIRF.py -h prints out the list of available options with a brief description of each.

Command-line options:
[-h]
[--ft features [features ...]]
[--nc n_cores]
[--eff Fix_eff [Fix_eff ...]]
[--nclass NCLASS]
[--train TRAIN]
[--train_format {txt,FITRES,csv}]
[--validation VALIDATION]
[--test TEST]
[--data DATA [DATA ...]]
[-spec SPEC]
[--alltypes_colname_spec ALLTYPES_COLNAME_SPEC]
[--phot PHOT]
[--spec_nofp SPEC_NOFP]
[--user_data USER_DATA [USER_DATA ...]]
[--user_alltypes_colnames USER_ALLTYPES_COLNAMES [USER_ALLTYPES_COLNAMES ...]]
[--filedir FILEDIR] [--store] [--restore]
[--use_filenames]
[--pklfile PKLFILE]
[--pklformat {txt,FITRES,csv}]
[--train_only] [--done_file DONE_FILE]
[--filestr FILESTR]
[--format FORMAT]
[--commit_hash_path COMMIT_HASH_PATH]
[--plots {'',Performance,SALT,Hubble,Error,Color,Magnitude,Bazin [{'',Performance,SALT,Hubble,Error,Color,Magnitude,Bazin} ...]]
[--weights WEIGHTS]
[--sim {Training,Validation} [{Training,Validation} ...]]
[--totals] [--save {Ia,CC,All} [{Ia,CC,All} ...]]
[--user_colors USER_COLORS [USER_COLORS ...]]
[--user_markers USER_MARKERS [USER_MARKERS ...]]
[--minmax]
[--debug]
[--alltypes_colname_train ALLTYPES_COLNAME_TRAIN]
[--alltypes_colname_validation ALLTYPES_COLNAME_VALIDATION]
[--alltypes_colname_test ALLTYPES_COLNAME_TEST]
[--type_labels TYPE_LABELS [TYPE_LABELS ...]]
[--type_values TYPE_VALUES [TYPE_VALUES ...]]
[--cv]
[--sample {t,v,b}]
[--niter {1,2,3,4,5,6,7,8,9,10}]
[--testsize TESTSIZE]
[--nfolds NFOLDS] [--pc]
[--purities PURITIES [PURITIES ...]]
[--withhold types/templates [types/templates ...]]
[--prvar]
[--Bazincuts {train,plots,off}]
[--noBazinparcuts]
[--Bazinpar_max BAZINPAR_MAX [BAZINPAR_MAX ...]]
[--Bazinpar_min BAZINPAR_MIN [BAZINPAR_MIN ...]]
[--Bazinerrcuts]
[--Bazinerr_max BAZINERR_MAX [BAZINERR_MAX ...]]
[--H0 H0]
[--OmegaM OMEGAM]

This script uses a machine learning (ML) algorithm to train a photometric classifier into classes
{Ia|CC or Ia|Ibc|II or I|II->Ia|Ibc}.
Select Random Forest features to use:

optional arguments: -h, --help show this help message and exit
--ft features [features ...]
Choose SN-type features to use for classification.
List them (space-separated) and select from: {c, x0, x1, t0, z, chi2, fit_pr, gpeak, rpeak, ipeak, zpeak, ra, dec, grpeak_m, ripeak_m, izpeak_m, grpeak_s,
ripeak_s, izpeak_s, Deltagr, Deltari, Deltaiz, snr1, snr2, snr3, t0_err, x0_err, x1_err, c_err,
Bazin_g_t_rise, Bazin_g_t_fall, Bazin_r_t_rise,
Bazin_r_t_fall, Bazin_i_t_rise, Bazin_i_t_fall,
Bazin_z_t_rise, Bazin_z_t_fall, grBazin, riBazin,
izBazin, CID, IDSURVEY, TYPE, FIELD, CUTFLAG_SNANA, zHEL, zHELERR, zCMB, zCMBERR, zHD, zHDERR, VPEC,
VPECERR, HOST_LOGMASS, HOST_LOGMASS_ERR, HOST_sSFR, HOST_sSFR_ERR, SNRMAX1, SNRMAX2, SNRMAX3, PKMJD,
PKMJDERR, x1, x1ERR, c, cERR, mB, mBERR, x0, x0ERR, COV_x1_c, COV_x1_x0, COV_c_x0, NDOF, FITCHI2, FITPROB, SIM_TYPE_INDEX, SIM_TEMPLATE_INDEX, SIM_LIBID,
SIM_NGEN_LIBID, SIM_ZCMB, SIM_VPEC, SIM_DLMAG,
SIM_PKMJD, SIM_SHAPE, SIM_AV, SIMNULL1, SIM_RV,
SIM_x0, SIM_mB, CID, zCMB, PKMJD, PKMJDERR, x0ERR,
x1ERR, cERR, FITCHI2, NDOF, MU, MUERR, FITPROB,
SNRMAX1, SNRMAX2, SNRMAX3, SIM_TYPE_INDEX,
SIM_TEMPLATE_INDEX, SIM_DLMAG, snid, z, t0,
['PKMJDerr', 't0_err'], chi2, fit_pr, mu, muerror,
snr1, snr2, snr3, sim_type, sim_nonIa, sim_mu, x0_err, x1_err, c_err} (default: ['fit_pr', 'x1'])
--nc n_cores Number of cores to use for parallelization (default: 4)
--eff Fix_eff [Fix_eff ...]
Efficiencies at which to evaluate purity (default:[0.95])
--nclass NCLASS Number of classes used in classifier (2=Ia/CC,
3=Ia/Ibc/II,-2=2x2 training: 2-stage--2-way
classification with I/II -> Ia/Ibc) (default: 3)
--train TRAIN Filename for training (default:
DES_training_SNR550.txt)
--train_format {txt,FITRES,csv}
Format of training data (default: txt)
--validation VALIDATION
Filename for validation; set to null to skip (default: '')
--test TEST Filename for test (skip if set to "") (default:
DES_test_SNR550.txt)
--data DATA [DATA ...]
Classify selected simulated/observed data
(default=Phot); choose from [Test, Spec, Spec_nofp,
Phot, ...] plots (if selected) will be normalized to
first data sample in list; customize user-data labels by adding user-supplied labels here (default: [])
--spec SPEC Filename for spectroscopic data (default:
specType_forcePhoto_bazin_v3.txt)
--alltypes_colname_spec ALLTYPES_COLNAME_SPEC
Column name for true SN types in Spec data (default:
spec_eval)
--phot PHOT Filename for photometric data (default:
host_prior_DESY1-4_bazin_v3e.txt)
--spec_nofp SPEC_NOFP
Filename for spectroscopic data without fit
probabilities (default: specSN_SNANA_nofpcut.txt)
--user_data USER_DATA [USER_DATA ...]
Filenames for user-supplied data; default labels in
plots etc. will be User0, User1, ... if no additional arguments supplied for --data (default: [])
--user_alltypes_colnames USER_ALLTYPES_COLNAMES [USER_ALLTYPES_COLNAMES ...] Column names for types of labeled user-supplied data
(unlabeled data assumed); use null string for
unlabeled data in list of user-data files with mixed
labeling properties. (default: [])
--filedir FILEDIR Directory name (relative or absolute) for storing
output files (default: ./)
--store Save trained classifier to pklfile (default: False)
--restore Restore trained classifier from pklfile (default:
False)
--use_filenames Use filenames as supplied (no default additions
specifying formats, etc. (default: False)
--pklfile PKLFILE Base filename for reading/writing trained classifier
({format}{nclass}way.pkl is auto-appended;
pre+filename also read/written if nclass == -2)
(default: trained_RFclassifier)
--pklformat {txt,FITRES,csv}
Format of stored training data (default: txt)
--train_only Run training only, save trained classifier to pklfile, and exit (default: False)
--done_file DONE_FILE
Path to file recording SUCCESS/FAILURE of run
(default: '')
--filestr FILESTR Choose string to append to filenames for output files (default: '')
--format FORMAT Format for output of classification data (txt, hdf5,
fits) (default: .txt)
--commit_hash_path COMMIT_HASH_PATH
Path to local github repository for retrieving commit hash (set to '' to skip) (default: ./ML-SN-Classifier) --plots {,Performance,SALT,Hubble,Error,Color,Magnitude,Bazin} [{,Performance,SALT,Hubble,Error,Color,Magnitude,Bazin} ...] Make selected groups of plots; choose from
Performance, SALT, Hubble, Color, Error, Magnitude,
Bazin; null argument supresses all plots (default: )
--weights WEIGHTS Name of data sample to use for computing weights for
plot normalizations; defaults to first data sample in
data list (default: )
--sim {Training,Validation} [{Training,Validation} ...]
Plot selected simulated data (default: ['Training'])
--totals Turn on totals in plots (default if using --data or
--user-data) (default: False)
--save {Ia,CC,All} [{Ia,CC,All} ...]
Types to save for printing pickling in plotting
modules (default: Ia)
--user_colors USER_COLORS [USER_COLORS ...]
Colors for user-supplied data (3 defaults available)
(default: ['seagreen', 'springgreen', 'lime'])
--user_markers USER_MARKERS [USER_MARKERS ...]
Markers for user-supplied data (3 defaults available)
(default: ['<', '>', '8'])
--minmax Turn on minmax print-out in plots (default: False)
--debug Turn on debug print-out in plots (default: False)
--alltypes_colname_train ALLTYPES_COLNAME_TRAIN
Column name for true SN types in validation data, eg.
"type3" identifies SN as Ia, Ibc, II; code checks for
consistency with --nclass and creates appropriate data
column(s) if needed (default: type3)
--alltypes_colname_validation ALLTYPES_COLNAME_VALIDATION
Column name for true SN types in validation data,
(default: type3)
--alltypes_colname_test ALLTYPES_COLNAME_TEST Column name for true SN types in test data, (default: type3) --type_labels TYPE_LABELS [TYPE_LABELS ...] Labels for classes in alltypes_colname (supply one per typename); CC assumed to be ~Ia (default: ['Ia', 'II', 'Ibc', 'Ib', 'Ic']) --type_values TYPE_VALUES [TYPE_VALUES ...] Corresponding class values in type_labels (default: [[0], [2], [1], [1], [1]]) --cv Turn on cross-validation (default: False) --sample {t,v,b} Sample to use for cross-validation (t=training, v=validation, b=both) (default: t) --niter {1,2,3,4,5,6,7,8,9,10} Number of ShuffleSplit cross-validation iterations (default: 5) --testsize TESTSIZE Validation sample size for ShuffleSplit (default: 0.5) --nfolds NFOLDS Number of folds for Kfold sample (default: 2) --pc Turn on purity comparison (default: False) --purities PURITIES [PURITIES ...] Purities for loop over test files (default: [0.1, 0.25, 0.5, 0.75, 0.9]) --withhold types/templates [types/templates ...] Hold-out test: withhold type (20=IIP, 21=IIN, 22=IIL, 32=Ib, 33=Ic) or template (206,209=IIN, 002=IIL, 201-235=IIP, 103-234=Ib, 021-218=Ic) from training sample: List them (space-separated) and select from: {20, 21, 22, 32, 33, 206, 209, 002, 201, 204, 208, 210, 213, 214, 215, 216, 219, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232, 233, 235, 400, 401, 402, 403, 103, 104, 105, 202, 203, 212, 234, 021, 022, 101, 102, 205, 207, 211, 217, 218, 502, 503, 506, 509} (default: ) --prvar Turn on probability-variance plots (default: False)
--Bazincuts {train,plots,off} Include Bazin selections in training/plots (TBD),
plots only, or turn off (default: off)
--noBazinparcuts Turn OFF Bazin parameter cuts (default: False)
--Bazinpar_max BAZINPAR_MAX [BAZINPAR_MAX ...]
Cuts on Bazin fit parameters: [A, t0, t_fall, t_rise, C] (default: [800, '', 150, 't_fall', 100])
--Bazinpar_min BAZINPAR_MIN [BAZINPAR_MIN ...]
Cuts on Bazin fit parameters: [A, t0, t_fall, t_rise,
C] (default: [-999, 1, 0, 0, -999])
--Bazinerrcuts Turn ON Bazin error cuts (default: False)
--Bazinerr_max BAZINERR_MAX [BAZINERR_MAX ...]
Cuts on Bazin fit errors: [A_err, t0_err, t_fall_err,
t_rise_err, C_err] (default: [100, 50, 100, 50, 100])
--H0 H0 Value of H0 (default: 68.62)
--OmegaM OMEGAM Value of OmegaM (default: 0.301)

ML_globals.py contains the definitions of global constants used throughout the code. The code classifies SNe in either binary (Ia or CC) or tertiary (Ia, Ibc, or II) mode, corresponding to the user choice --nclass 2 or --nclass 3, respectively.

Two formats are supported for the user-supplied input data, which are read in as astropy tables:

1. text file with a header that labels columns of data,
2. fitres file (typically output from SNANA).

ML_globals.py documents the default formats for these files. User options are available to change some of these default values (see above). The most important definitions are:

**allowed_features: dictionary with list of allowed feature names for each allowed format
data_defaults: dictionary with information about how classes are labeled for each allowed format
allowed_templates: dictionary with list of templates used to simulate each class in the data

The contents of allowed_features and allowed_templates should be self-explanatory. The following keys in the data_defaults dictionary describe the labels used for the various classes contained in user-supplied labeled data. (These defaults can be changed on the fly).

alltypes_colname: name of data column containing the labels for the class (type) of each object
type_values: value (usually integer) given to each labeled class
type_labels: labels (Ia, Ibc, II, CC) corresponding to each value in type_values

The information in the following keys is used internally in the code. type_colnames: name of column containing labels for different values of nclass (2, 3, -2);
generated on the fly (if possible) if not available in data file
alltypes_available: True (False), if labels (do not) differentiate between classes of CC SNe

In general, if your file format/variable names are different from those listed in the above dictionaries, you will need to make changes to ML_globals.py to get the code to run successfully. New feature names should be added to the list appropriate to your file format in allowed_features. Although it is possible to change the defaults in data_defaults by supplying user options, it will be more convenient to make the appropriate changes to the data_defaults dictionary by modifying the entries for the keys alltypes_colname and type_values to match your labeled data.