- UC 16 - make it ready for e-mail annoucement
- remove unused branches - FSPL_no_Bozza, some other ones?
- check notes below
- are we using VBBL with improvements from 2018 paper?
- code not yet well documented - see below
- print Model/ModelParameters: add t_0_par somewhere - maybe only when it is != t_0
- chi2 gradient for finite source effects. (requested by JCY)
- "import MulensModel as mm" everywhere: binarylens.py pointlens.py use_cases/->30,31,34 examples/run_time_tests/check_system.py examples/checks.py TUTORIALS
boldfaced tasks are most important because requested by the users
italics mark important tasks
Changes for planned v2 are here: documents/MM_v2.md
- Install
- PYPI website - some links: 1 2
- test Windows installation
- in setup.py in setup() add keywords: long_description, classifiers
- virtualenv; pip install -r requirements.txt; its best to install the dependencies first
- more metadata in setup.py
- more on setup.py: link
- compile with "pedantic" flags for compilers
- Documentation
- Sagan workshop hands-on activity in MM
- examples as ipython notebooks
- Add __repr__ functions to Lens and Source
- code not yet well documented - some should changed to private in V2.0.0:
- RA & Dec in coordinates.py (maybe also code it better)
- Horizons.file_properties
- MulensData.plot_properties
- SatelliteSkyCoord.ephemerides_file
- Trajectory.earth_coords - not implemented, so maybe remove?
- Include full documentation via setup.py data_files mechanism.
- note that all plotting functions require plt.show() or plt.save()
- try removing Attributes from docstrings - just make short @property functions
- add a note that pi_E is "geocentric" (and "heliocentric" has the same length of vector but is rotated)
- example 8 corrections - PSBL, not PSPL; clarify removing the anomaly
- make sure that website shows correct version of MM
- note that we are not checking for negative source or blending flux
- add a list of public datasets: VVV paper was published?, add link to K2/MCPM?; LINK the file documents/public_data_list.md somewhere
- Effects:
- Binary source - see documents/binary_source_notes.md:
- extract flux ratio for binary source models when fluxes are fitted via regression
- finish use cases
- source_flux_ratio added to ModelParameters
- Fit.fit_fluxes docstring to be updated
- which_parameters() - note that it doesnt work for binary source parameters, but the parameters work properly; just BSPL and rho_2 etc. are optional
- models with fixed no blending: fit_blending keyword changes
- parallax models
- binary source - there should be one Fit less in Event.get_chi2xxx functions - if there are 2 sources, then calculate magnification and use F_S = F_S_1+F_S_2 and get it from self.model.fit; most probably adding some function to Fit would help
- test binary-lens binary-source
- different t_E for each source (correct Model.set_times)
- test binary source with exactly one rho_X defined
- add t_eff_1, t_eff_2
- Magnification function provided by the user - already started in user_method branch; also this could be used to model variable source events - note that
- triple lens
- test ob07349 to see if center of mass can reference point for t0 and u0
- caustics calculations
- plan use cases and unit tests - note in documents/TRIPLE_LENS.md
- "reset" triple_lens branch? (i.e., copy code, remove branch, start new branch e.g. tripleLens, use the code copied at begin)
- documents/TRIPLE_LENS.md - make notes in order there
- check triple lens use cases in master branch
- use cases, point source, hexadecapole...
- triple source calculations
- Finite Source
- FSPL with low magnification - do Witt & Mao 94 or Witt 95 give the right formulas?
- FSPL ray shooting (ala getmag_rs_single.f)
- Yoo+04 full formalism
- get gamma/u LD coeffs from Claret papers etc.
- Lee+09 - gradient calculations for uniform source, also faster calculations - profile
- FSPL for large sources using Agol 2003
- Xallarap (see below for references)
- Quadratic limb darkening
- Multi-lens ray shooting:
- mapmaking version which adds new rays as needed (but remember that it runs for fixed (s,q) only!)
- Yossi idea to find all the images
- Orbital motion like in VBBL 2.0
- calculate jerk parallax degeneracy: Park+04 Gould 04
- topocentric/Earth parallax
- Chang-Refsdal binary calculations
- elliptical source magnification Heyrovsky & Loeb 1997
- magnification calculated for a set of points, not just a trajectory - this way we could, e.g., plot magnification maps
- fit_blending for only some of the datasets
- blending flux fixed at user-specified value - requires use case
- Binary source - see documents/binary_source_notes.md:
- Parameterization
- Cassan 2008 binary lens parameters:
- option to change scaling (from [0,1] to C08 params) to work well near topology change
- Albrow et al. 1999 (also Cassan 2008 Sec. 5)
- t_eff as a parameter for large u_0 - see Andy's paper and maybe also other from Jen's 2012 paper, i.e., f_lim=f_s/u_0 and q*t_E
- caustic size w Dong+09 refers to Chung+05; see also Skowron+11 Eq.2
- check if new parameters are defined here: Liebig, D'Ago, Bozza, and Dominik 2015
- Heyrovsky 2003 parametrization of limb-darkening
- t_0_planet, u_0_planet, t_E_planet instead of s, q, alpha
- Dominik 2009 for PSPL
- Jung+17 - rotating triple lens - somehow special version of xallarap
- Cassan 2008 binary lens parameters:
- Function Improvements/Expansion:
- BinaryLens class:
- VBBL2.0 - are we using accuracy limit as default? If so then we should switch to relative accuracy
- should BinaryLens() accept source_x/y as lists or arrays?
- function for center of mass shift (currently: shift_x in trajectory.py, x_shift in binarylens.py, xcm_offset in caustics.py)
- central and planetary caustic properties: Chung et al. 2005 and Han 2006
- consider using Utils.complex_fsum() in BinaryLens functions: _jacobian_determinant_ok_WM95()
- faster hexadecapole using Cassan 2017 (code)
- there should be an option for the user to ignore "wrong number of solutions" error and replace it with a warning
- there should be a way to change the "wrong number of solutions" error into warning or dismiss it fully
- error message for not 3 or 5 images - give diagnostics first
- error message for not 3 or 5 images - This can be improved by checking the denominators in the equation (before polynomial is fully formed - see Zoey's Mathematica notes)
- error message for not 3 or 5 images - in "Consider ..." note that one must use Model.set_magnification_methods() for epochs far from peak
- VBBL additional parameter so that it works for huge source and tiny caustic
- Caustics class:
- Caustics._calculate - optimize using vectors instead of a loop
- Caustics calculations using Erdl & Schneider 1993 approach; there is also Asada 2002 and I don't know if it makes sense to code that one as well
- root solver can be used the same as in binarylens.py - not needed for binary lens and Erdl & Schneider 1993
- smaller points
- correct conditions in get_caustics()
- Event class:
- Allow fluxes to be fixed in chi^2 calculation (e.g. given a particular fs, fb, which you might want to do if you want fs as a chain parameter); also think how it will work for binary sources
- give access to all fluxes without changing data_ref
- plot magnitude difference between 2 models for residuals plot
- it seems it doesn't have plot_trajectory()
- add plot_source() etc. already implemented in Model
- Event should sync information on which of the 3 types of parallax are used, so that if it's specified for event, then there will be exception if one dataset is missing earth_coords etc. In general there should be some way to make sure which parallax types are used in which calculation of magnification.
- Class Event should have not only set_datasets() methods but also add_datasets(), i.e. a similar method that appends datasets to self._datasets.
- reduce calls to Fit.fit_fluxes()
- add finite source in chi2_gradient()
- chi2_gradient() should cope NaN values in a way similar to get_chi2()
- check all functions that should pass fit_blending parameter - Event.plot_model, what else??? Already done: Event.get_ref_fluxes()
- chi2 with maximum value provided - if the chi2 for point-source gives chi2 larger than specified limit, then finite source calculations are not undertaken (this should significantly speed-up MultiNest)
- get flux and its error in reference system
- change order to improve the website
- gradient - fluxes as well? if so, then start using the second test in test_event_chi2_gradient()
- for consistency, it would be good to combine get_chi2_for_dataset() and get_chi2_per_point()
- other likelihoods, e.g., Dominik+18, ARTEMiS, SIGNALMEN, RoboTAP
- function that calculates cumulative chi2 so that it can be plotted easier
- Binary source - see optimization comment at begin of Event.get_chi2_for_dataset()
- plot cumulative chi2 difference between 2 models - magnification and magnitude spaces
- Fit class:
- should use marginalized distributions of fluxes (if those are from linear fits); JCY - it needs UC
- n_sources somehow inconsistent in different places
- Horizons class:
- JPL Horizons
- correct JPL Horizons => CSV file format; also example usage
- check if Horizons e-mail is for correct satellite
- BJD
- conversions to BJD from HJD, JD etc. (astropy link)
- BJD_TDB in satellite ephemeris astropy link
- JPL Horizons
- Lens class:
- __repr__ function needs work
- a_proj, couples with source distance in mulensmodel to determine s.
- 2-body example 3 is missing s. Why? Does that work?
- problem with tracking number of masses, esp when successively defining masses (see test_Lens.py)
- implement triple+ systems
- MagnificationCurve class:
- re-write magnification() to use lazy loading (here or in model.py)
- Model class:
- in functions magnification(), plot_magnification(), and plot_trajectory() use satellite_skycoord from __init__ if available
- plot_lc() - add satellite option like in plot_magnification(), other options as well - use keywords passed to self.magnification()
- reorder functions so that it looks good on website
- Model.set_parameters() should remember previously set values (of course unless they're overwritten)
- Class Model should not allow accessing attributes that shouldn't be there, eg., q for single lens case.
- Function that prints RA, Dec, t_0_par, t_0_kep, types of parallaxes turned on, satellite info, limb coeffs and textual description of type of model
- plot_trajectory() - mark epochs using colorscale? Maybe it should be passed by kwargs (if so, then add example)
- Should get_satellite_coords() use caching?
- we should have versions of all plot functions to use magnifications instead of magnitudes; also add access via Event
- add option to use random zorder when plotting multiple datasets (e.g. gaussian with sigma depending on number of plotted datapoints)
- does Model.plot_lc() give docstring for flux_ratio_constraint option? If not, then present it in plot_magnification() docstring
- use plt.axhline() to plot 0 line in residuals plots at the end, using t_min,t_max
- get_residuals needs unit test
- plot_trajectory - mark data epochs (as pyplot kwargs use MulensData.plot_properties)
- plot_data & plot_residuals - if color is not set by the user and show_bad is True, then X-s are plotted using different color
- allow rotating plots, so that source travel "exactly" from left to right
- when plotting residuals allow keywords to be passed for plotting 0-line
- get_data_magnification() - should there be data.good masking?
- get_ref_fluxes - use caching (self._fit) inside it
- plot_lc() should have satellite_skycoord keyword as plot_magnification() has
- plot_source and binary sources - allow different kwargs for each of the sources?
- check_if_caustic_crossing() function can be added - for non-parallax, no-orbital motion it requires calculation of cusps positions for each caustic and checking if they are on the same side of the trajectory
- try/except in pyplot commands and catch AttributeError - these may be miss-spelling etc.
- ModelParameters class:
- values in dimensionless astropy.quantity should be changed to float, other types should be rejected (unless it's a time unit etc.)
- LaTeX strings with parameters names (useful e.g. for corner plots or tables with results)
- Transform t_E and other parameters between geocentric and heliocentric frames.
- option to return alpha, dalpha_dt, and ds_dt as floats instead of astropy.quantities
- why .rho returns None if it's not defined? In other similar cases we have KeyError. Should that be changed? (if so, then maybe only after changing version to 2.0.0)
- to make _check_valid_combination_1_source_...() shorter, make a boolean dict that says if given parameter is defined or not
- change order to improve the website
- check if t_eff and t_star can be used as input simultaneously
- check if input values are floats (or something else accepted)
- warning if t_0_par is changed too many times
- add is_Cassan08 or is_standard_binary?
- __repr__ - if Cassan 2008 then should we print standard parameters as well?
- MulensData class:
- Errorbar scaling, in particular the two parameter.
- quick look alignment for MulensData objects - just use interpolation
- make __init__() shorter
- add version of n_epochs that uses only good epochs
- read settings from file header: flux vs. mag, filter, satellite info
- change order to improve the website
- docstring phot_fmt vs. input_fmt
- data_and_err_in_input_fmt() and Fit.get_input_format() possible can be deprecated or removed because we shifted to chi2_fmt instead of input_fmt
- when plotting data, make sure that max/min limits on Y axis include errorbars, if the errorbars are shown
- export/save given data file in scale of other dataset and model
- this line may be wrong for some values of char: kwargs['fmt'] = kwargs['fmt'].replace(char, "")
- for plt.scatter() the color can be set as 'facecolor', 'facecolors', or 'edgecolors' and this should be dealt with in _set_plot_properties()
- for plotting X for bad data use large size and/or thinner line
- separate colors (or even kwargs) for X-es as an option (to get contrasting colors see https://itsphbytes.wordpress.com/2016/08/29/complementary-colors-python-code/)
- PointLens class:
- make WM method faster: 1) interpolation done once for many values; 2) interpolate different function; 3) allow changing number of annuli; 4) divide A by 2 different functions (z<1 or z>1) and interpolate these - like in VBBL (arguments of interpolation are: 1) (rho, z), 2) (log rho, log z) etc.)
- add Witt and Atrio-Barandela 2019?
- get_pspl_magnification() - change it to operate on u^2, not u, so that np.sqrt() calls are reduced
- 1+2/u^4 approximation for very large u
- try to remove sympy and use scipy instead (not possible in Sep 2020, because there is no elliptical integral of third kind in scipy)
- SatelliteSkyCoord class:
- attach magnification_methods to SatelliteSkyCoord so that they overwrite Model and MagnificationCurve settings when given SatelliteSkyCoord is used
- Trajectory class:
- warning when too many annual parallax calculations are conducted
- _get_delta_satellite() should be using self.times
- annual parallax caching - if moved to MulensData, then would be faster because hashing of coords and time vector takes time
- maybe Trajectory should be able to plot itself, and Model.plot_trajectory() should call it - it would be easier for binary sources etc.
- colorscale time or magnification (see Fig 2 in Ranc+19)
- plot in units of theta_star (or even days) instead of theta_E
- UniformCausticSampling class:
- add function that calculated t_in & t_out in analogy to get_x_in_x_out() - the problem is that for given (u_0, alpha) - there are 2, 4, or 6 caustic crossing points
- prior probability (in get_uniform_sampling()) of hitting given caustic based on (d, q, alpha) - see Kains+12
- prior only in given box (in get_uniform_sampling())
- prior includes t_E prior provided by user (also add option to use Mroz+19 etc) in get_uniform_sampling()
- trajectories that cross 2 different caustics
- profile calculations
- Utils class:
- in np.any() ifs give more information in warning e.g., "out of 1234 values provided, the fails are: 12, 345, 678 (0-based)"
- add u(a) function: u = np.sqrt(2A/np.sqrt(A^2-1.) - 2.)
- utils.py:57 - code produces 2 warnings, should produce just one; use masking
- documentation - use ITALICS
- uses masks there if warnings and apply None etc. - check examples after change!
- add a function that calculates mag/flux in a different flux space - we currently do it in MulensData.plot() and Model.get_residuals()
- MulensObjects submodule:
- Plotting:
- plt.plot() .scatter() and .errorbar() should share default colors, so that when you plot one dataset and one model and don't set colors, then they are of different colors
- for plotting functions option to pass pyplot.Axis and pyplot.Figure instances and call e.g. Axis.scatter() instead of pyplot.scatter(); for a simple example see here
- subplots with shared X-axis (plt.subplots(2, 1, sharex=True, gridspec_kw={'height_ratios': [4, 1]}, figsize=???, dpi=100)) - start in Example 5
- add option to plot satellite coordinates as in Henderson+16 where K2 and Spitzer orbits were compared; i.e., make Trajectory._get_delta_satellite() public and add appropriate example
- add plotting with fit_blending=False for functions that use magnitude space
- add plt.xlim() and ylim in plotting functions (using t_start) etc.; then also update (simplify) tutorials, examples etc.
- caustics for trajectory plot with single lens models
- plot upper limits instead of photometry with negative flux
- Examples:
- Hamiltonian MCMC link 1 and link 2 and link 3
- example with parallel EMCEE
- plot many models from posterior
- chi2 per dataset
- scipy.curve_fit() and print parameter uncertainties
- corner plots; they require corner, pyhdust, or pyGTC
- F_s for MOA data for MB08310 differs from Janczak paper - is it caused by FSPL vs. FSBL models?
- add example that shows 'log_' in the name of the parameter; central caustic anomaly planet would be best,
- example 08: PSBL, not PSPL
- gaussian process - see Li+19 (maybe it requires input in Fit class, hence would not be just an example)
- Dan FM approach for periodic variables - two variables instead of one (identified by some postfix) - make it for alpha or x_caustic_in/out
- add illustration on how to remove airmass trends
- add example of fitting PSPL model using Albrow (2004) method link
- plotting - sharex where possible
- note in PSPL tutorial about plotting data in MulensData
- add example that after calling Event.get_chi2() use Event.fit to get e.g. magnifications so that the magnification is not calculated twice
- satellite data fitted and plotted - what is missing now?
- use ast.literal_eval() for .cfg files to read dict or list, e.g., for MulensData options.
- high-level fitting example where we don't care how complicated it is, we just want to make it simple and useful for the user
- some cfg files use "../data/..." - change it to MM.DATA_PATH somehow
- in emcee we should check if all the starting points are in prior
- check if MM correctly interacts with scipy.optimize.leastsq and maybe add an example
- add Coordinates.velocity_of_Earth() example
- Example 13 - make x_caustic_in/_out periodic variables
- add example with well-known code like https://mc-stan.org/
- MulensData.bad needs example note that one has to substitute full vector, not single values
- Miscellaneous:
- COVERAGE : "coverage run --source MulensModel -m py.test" and then "coverage report" or "coverage report -m" or "coverage html" (and then open htmlcov/index.html); https://coverage.readthedocs.io/en/v4.5.x/
- u_0 sign for satellite or just parallax model - some way of following u(t) evolution
- warnings - give type for each one of them
- Dave's idea on caustic-crossing epochs as parameters t_cc1 and t_cc2 - see david.p.bennettATnasa.gov e-mail on Feb 11, 2019
- when checking units use Unit.physical_type - search for physical_type in mulensobjects/lens.py as an example; to find places to be changed search for "isinstance" (to find these places run grep isinstance *py mulensobjects/*py | grep Quantity
- use lazy loading in MagnificationCurve.magnification and/or Model.magnification
- guessing parameters of PSPL model (Kim+17 as an example)
- add calculation of Caustic Region of Influence (CROIN) - Penny 2014
- anything from use cases that does not work yet -- see TODO.md file
- comments at begin of each use case and example
- use case 25 doesn't work
- OB161195 is used only in UC25 - change it and remove directory
- interaction with fitting routines - see list of them
- caching of results in trajectory.py should stop at some point - if the user changes t_0_par or coords, then there is no point in remembering huge indexes (whole self.times)
- profile the code (python -m cProfile script.py - you only should check 2nd and 4th column)
- Leap seconds library - barycorrpy
- documents/TODO.md file - move content and remove
- add transformation Jacobians, see: Batista+11 and Skowron+11
- use cython, numba, pypy or similar (numba seems best) to speed-up calculations in Lee+09 and/or Cassan08
- remove "import *" from __init__.py and mulensobjects/__init__.py
- BinaryLens class:
- Other Tests:
- test_event_chi2_gradient() - add parallax without flux gradient
- add unit tests for Horizons and MulensData.satellite_skycoord
- Coordinates - write tests, possibly remove test_Coords.py
- t_eff is not tested
- plt.scatter -> plt.plot; after that we can start unit tests for plt calls
- check for memory leaks by running long calculations and monitoring RAM usage
- Style/Architecture:
- Are we consistent with PEP8? check here or pycodestyle command
- PEP8 for tests/ (test_Event is already done: test_Event test_ModelParameters test_Model_Parallax
- for examples/
- better import of the module so that all main classes are accessible (use __all__ = [...] in all files?)
- Utils - Make subpackage/submodules that group related functions (e.g. flux2mag conversions)?
1st tutorial, 2nd tutorial, example
Griest & Hu 1992, Han & Gould 1997, Dominik 1998, ob9919 - Smith et al. 2002, Ghosh et al. 2004, Jiang et al. 2004, Dong et al. 2009, ob07514 - Miyake+12, mb10328 - Furusawa et al. 2013, ob130911 - Miyazaki et al. 2019, Poindexter et al. 2005 - 23% of events are affected by xallarap, ob07368 - Sumi et al. 2010 and Suzuki et al. 2016, ob150845 = mb15277 - Calen leads, Roman predictions - Miyazaki+20,