TUTORIAL

Step 3. Classification

$ vi classify_baygaud_main.hanning_regridded.cube.py

You should set the environment global variables below.

• n_gauss : the number we input in the generate_baygaud_segs_scripts process. e.g., 2, 3, 4, ....

• _cube.name : input the cube.fits name e.g., galaxy_name.cube_Vel.fits

• bulk_ref_vf : input the cube_ref_vf name e.g., galaxy_name.moment1.fits

• n_sub = 11 : # of sub-files

• sn_limit = 2 : the SN of a profile should be larger than this value.

• channel_width = 3.97 # [km/s] : you can check this by reading the fits's header

CDEL3 = 3974.09

CUNIT3 = m/s

In this case, the channel_width will be 3.97 [km/s]

• vel_bulk_limit = 5 # [km/s] : If the variation is larger than 5 , the gas component is regarded as non-circular gas component. As the number is smaller, we can extract the closer components to bulk components.

• vdisp_cnm = 4.0 # [km/s] : if the vel_dispersion is smaller than this value, the gas is regarded as cold neutral medium gas.

• vdisp_wnm = 8 # [km/s] : if the vel_dispersion is larger than this value, the gas is regarded as warm neutral medium gas.

• vdisp_hot = 99 # [km/s] : Set the large number.. If the vel_dispersion is larger than this value, it is hot gas.

If the setting is done, let's proceed on the classification step.

$ python classify_baygaud_main.hanning_regridded.cube.py

Step #. Generating segments, running BAYGAUD, and merging segments

# added by hj

Before you run baygaudGUI.py, install all required dependencies:

• python3-tk

• libbz2-dev

• numpy

• matplotlib

• glob

• astropy

• spectral_cube

• natsort

• tqdm

• fitsio

# ms

Firstly, we are going to cut the cube into parts, which we will call them 'segments'.

Doing so, progress can be saved if there goes something wrong in the middle of the computing process.

This can be done with 'generate_baygaud_segs_scripts.cusped0'.

Open it with any text editor you have, for example;

• code generate_baygaud_segs_scripts.cusped0

At the very first part, you'll be able to see

set main_baygaud =

set ncolm_per_core =

set nsegments_nax2 =

set n_cores_total =

set sleep_time =

set bayes_factor_limit =

main_baygaud should be given with the path where script baygaud is.

It should be at <baygaud directory>/src/bin/baygaud

ncolm_per_core is the horizontal pixel length of the segment, and also the amount of core that'll be used per segment.

So if the given cube has it's NAXIS1 = 600 and you give ncolm_per_core = 2, it will make 300 divisions in horizontal direction.

nsegments_nax2 is the number of divisions in the vertical direction.

If you give it 4, there will be 4 divisions in the vertical direction.

Continuing from the example in ncolm_per_core, if the given cube has its' NAXIS1=NAXIS2=600, and you give ncolm_per_core = 2 with nsegments_nax2 = 2, there will be 2 divisions in the vertical direction and 300 divisions in horizontal direction, resulting total 600 segments to run.

n_cores_total controls how many segments to be run at the same time.

It designates one core per a segment, so giving number means the amount of segments to be run at a same time.

sleep_time is sleeping time between running segments in second unit.

Note that it starts to count from the moment you run a segment, so if a segment is done later than given sleep_time it will add another segment to calculate before the already-running-one completed, which may lead to overload the cores.

It is recommended to set this value after try running a segment to figure out how much time it take to finish for a single segment.

bayes_factor_limit

In this example, we've set like below;

set main_baygaud = ~/lib/baygaud.v1.3.0/src/bin/baygaud

set ncolm_per_core = 10

set nsegments_nax2 = 2

set n_cores_total = 12

set sleep_time = 50

set bayes_factor_limit = 3.2

In the next part, you can set MultiNest parameters.

Setting as below should be ok but if you want more details, see Feroz et al. (2008) and/or https://johannesbuchner.github.io/PyMultiNest/pymultinest_run.html

set int = 1

set const_eff_mode = 1

set nlive = 100

set mmodal = 0

set efr = 0.9

set tol = 0.3

set feedback = 0

set write_multinest_outputfile = 1

set max_iter = 0

Below the multinest part, you'll be able to see

set nax1_s0 =

set nax1_e0 =

set nax2_s0 =

set nax2_e0 =

The galaxy may not be filling entire region of the cube, or you may want to run some specified zone in the cube.

In this case you can set upper values to specify a zone to run.

Each s0 and e0 should be given with starting pixel and ending pixel respectively.

For example, if the cube has NAXIS1=600 and you give nax1_s0 = 200, and nax1_e0=400, BAYGAUD will only run from 200 to 400 pixels in AXIS1 direction.

In this example, we'll set as

set nax1_s0 = 20

set nax1_e0 = 80

set nax2_s0 = 7

set nax2_e0 = 80

Which will only use the zone in the figure below

• ncolm_per_core: number of columns to distribute to one column. (e.g., RA)

• nsegments_nax2: number of segments in axis 2 (e.g., DEC)

• n_cores_total: number of total cores to run BAYGAUD

• sleep_time: break time after BAYGAUD runs N segments (N=n_cores_total / ncolm_per_core) in seconds

• nax1_s0: set number of pixels you will consider to make background for ax1 (e.g., RA) -> from the first index to nax1_s0

• nax2_s0: set number of pixels you will consider to make background for ax2 (e.g., DEC) -> from the first index to nax2_s0

• profile_sn_limit: cutoff limit of signal to noise (2~3 generally)

• bayes_factor_limit: when BAYGAUD finds more reliable optimal gaussian number, it uses Bayes factor for the model selection. This value is the evidence of the optimal gaussian number over the evidence of null gaussian number.

(한 픽셀에서 optimal gaussian number를 결정할 때, model selection을 하게 되는데, 이 때 선택되는 optimal gaussian number는 바로 다음으로 높은 evidence를 가지는 gaussian component number보다 적어도 3.2배 커야한다(?))