Dfisher_2022A Documentation
This project is being developed in the course of delivering the DFisher_2022A ADACS Merit Allocation Program project.
Installation
Pre-requirement:
python >=3.8 <3.10
HDF5 >= 1.8.4 (>=1.8.15 is strongly recommended)
Latest PyPI release
$ pip install dfisher_2022a
Common troubleshooting: If installation fails, try to upgrade pip
by running pip install --upgrade pip
first.
Latest dev-version on GitHub
$ pip install git+https://github.com/ADACS-Australia/dfisher_2022a.git#egg=dfisher_2022a
NOTICE: In the dev-version, a faster version of lmfit
(light-lmfit-py) is used. This version provides a fitting method, “fast_leastsq” in addition to other fitting methods available in lmfit(1.0.3)
. This method can be 2x faster than leastsq
. Check dev notes for more details.
Getting Started
Import the package
>>> import dfisher_2022a
Read in data cube
>>> cube = dfisher_2022a.ReadCubeFile("single_gaussian_muse_size.fits").cube
If a separate variance file is provide:
>>> cube = dfisher_2022a.ReadCubeFile("single_gaussian_muse_size.fits", "muse_var.fits").cube
Prepare data for fitting
>>> p = dfisher_2022a.ProcessedCube(cube, z=0.009, snr_threshold=5.)
1. De-redshift the cube
>>> p.de_redshift()
2. Select fitting region for a given line
>>> p.select_region("Halpha", left=20, right=20)
Keywords left
and right
set the wavelength cuts around the given line on both sides, e.g. the selected region is [line-left, line+right]. If this region exceeds the cube wavelength range, a nearest value within the cube will be used instead.
3. Filter the cube by SNR threshold
>>> p.get_snrmap()
Select fitting model
>>> model = dfisher_2022a.Lm_Const_1GaussModel
A single Gaussian model is available within this package.
Fit the cube
>>> cfl = dfisher_2022a.CubeFitterLM(data=p.data, weight=p.weight, x=p.x, model=model, method='leastsq') # accept lmfit.Model.fit kwargs
>>> cfl.fit_cube()
Additional keyword arguments for lmfit.Model.fit can be passed to the class object as well.
Save output
>>> out = dfisher_2022a.ResultLM()
>>> out.get_output(p) # get attributes from ProcessedCube object
>>> out.get_output(cfl)
>>> out.save()
An out
directory will be generated in the current directory.
Read output
In the .out
folder:
result.h5
fitdata/
where result.h5
stores the fitting result, and fitdata/
contains processed data used for fitting.
To read result.h5
file:
>>> import pandas as pd
>>> store = pd.HDFStore("result.h5")
>>> store.keys()
['/Halpha_Const_1GaussModel']
>>> df = store.get("Halpha_Const_1GaussModel")
Check available lines
>>> dfisher_2022a.EmissionLines
{'Halpha': 6562.819, 'Hb4861': 4861.333, 'Hdelta': 4101.742, ...
The line information is included in emission_lines.py
. Users can customize this file (e.g. adding more lines or updating the wavelength) before importing this package.
A wrapped approach
A wrapper function encapsulating steps 1-6 is available:
>>> from dfisher_2022a import fit_lm
>>> model = dfisher_2022a.Lm_Const_1GaussModel
>>> fit_lm(cubefile="single_gaussian_muse_size.fits", line="Halpha", model=model, z=0.009, left=20, right=20, snr_threshold=5.)
Use the faster version of lmfit
If dev-version of this package is installed, which uses a faster version of lmfit
as dependency, a faster fitting method is also available, by using method="fast_leastsq"
and adding an argument fast=True
>>> cfl = dfisher_2022a.CubeFitterLM(data=p.data, weight=p.weight, x=p.x, model=model, method='fast_leastsq', fast=True) # accept lmfit.Model.fit kwargs
>>> cfl.fit_cube()
In the wrapper function:
>>> fit_lm(cubefile="single_gaussian_muse_size.fits", line="Halpha", model=model, z=0.009, left=20, right=20, snr_threshold=5., method="fast_leastsq", fast=True)
Create custom model
Users can create their own models following the descriptions provided by lmfit. To use fast_leastsq
method in the dev version, eval_fast
needs to be written as a method of the model. See dev notes of light-lmfit-py for more details.
Special notes for using OzSTAR
The default number of worker processes used in the parallel fitting is os.cpu_count()
. In OzSTAR, it’s the total number of cores in one node, which is 36. It can result in runtime error if you also setup cpus-per-task
, e.g. cpus-per-task=4
in the sinteractive
mode or job submission scripts. In this case, please always specify the value of nprocess
when calling fit_lm
function or instantiating CubeFitterLM
class.