Update/change an existing target

You may want to change pre-defined skysurvey targets, either by changing the model function parameters (the kwargs) or even by changing the model itself.

You can do that while calling draw(), from_draw() or simply using update_model() or update_model_parameters().

Let’s see the differences.

tldr; kwargs in draw() only affect this data draw, while changes calling from_draw(), update_model() or update_model_parameters() affect the instance’s model.

---

Changing model entry’s parameters

This is the simplest and the most commun usage.

You want to change the model’s parameters. You can do that using **kwargs of draw() (from_draw()) method, with the format:

{entry_name: dict}

For instance, let’s change the alpha and beta parameters of the SNeIa model ; these are parameters of the magabs function (see, skysurvey.target.snia.SNeIaMagnitude.tripp1998):

import skysurvey

snia = skysurvey.SNeIa.from_draw(size=1_000, 
                                 magabs={"alpha":-0.15, "beta":3.2} # kwargs here
                                )
snia.data

	z	x1	c	t0	ra	dec	magabs	mwebv	magobs	x0	template
0	0.12335	0.005	-0.059337	56050.554688	295.345276	-45.356987	-19.578701	0.074459	19.297249	0.000303	salt2
1	0.12425	-0.515	0.185660	56089.160156	10.793458	-48.294235	-18.340822	0.014034	20.552164	0.000095	salt2
2	0.17995	0.130	-0.025264	56175.203125	330.346680	-55.670135	-19.534475	0.022090	20.237242	0.000128	salt2
3	0.15305	0.275	-0.075212	56125.285156	337.899323	-55.291012	-19.576645	0.020682	19.808245	0.000189	salt2
4	0.18125	-0.385	0.064151	56080.679688	185.410629	86.832001	-19.074341	0.293091	20.714678	0.000082	salt2
...	...	...	...	...	...	...	...	...	...	...	...
995	0.13895	-1.495	0.047259	56121.378906	318.926514	49.665226	-18.777458	1.906101	20.378557	0.000112	salt2
996	0.18115	0.065	0.230396	56017.242188	180.759109	-61.256058	-18.490904	2.337669	21.296787	0.000048	salt2
997	0.18155	-0.770	0.028307	56062.605469	26.737309	-17.553894	-19.163115	0.017458	20.629879	0.000089	salt2
998	0.08145	1.075	0.356473	56155.152344	96.943260	52.802578	-18.330139	0.132881	19.584503	0.000233	salt2
999	0.08725	1.270	-0.026995	56116.898438	155.215439	-21.428831	-19.558033	0.077002	18.514513	0.000624	salt2

1000 rows × 11 columns

Using kwargs in from_draw() does update the actual instance’s model, unlike the kwargs option of draw()

Consequently, the model itself is changed.

snia.model

{'redshift': {'kwargs': {'zmax': 0.2}, 'as': 'z'},
 'x1': {'func': <function SNeIaStretch.nicolas2021 at 0x2a1033640>},
 'c': {'func': <function SNeIaColor.color_rvs at 0x2a10329e0>},
 't0': {'func': <built-in method uniform of numpy.random.mtrand.RandomState object at 0x1100abe40>,
        'kwargs': {'low': 56000, 'high': 56200}},
 'magabs': {'func': <function SNeIaMagnitude.tripp1998 at 0x2a10336d0>,
            'kwargs': {'x1': '@x1',
                       'c': '@c',
                       'mabs': -19.3,
                       'sigmaint': 0.1,
                       'alpha': -0.15,
                       'beta': 3.2}},
 'magobs': {'func': 'magabs_to_magobs',
            'kwargs': {'z': '@z', 'magabs': '@magabs'}},
 'x0': {'func': 'magobs_to_amplitude',
        'kwargs': {'magobs': '@magobs', 'param_name': 'x0'}},
 'radec': {'func': <function random_radec at 0x17b8f0dc0>,
           'kwargs': {},
           'as': ['ra', 'dec']},
 'mwebv': {'func': <function get_mwebv at 0x2a1032d40>,
           'kwargs': {'ra': '@ra', 'dec': '@dec'}}}

How to permanently change a loaded instance’s model parameters ?

To permanently change the model of an already loaded instance, you need to use self.update_model_parameter().

snia.update_model_parameter(magabs={"alpha":-0.15, "beta":3.8} )
snia.model

{'redshift': {'kwargs': {'zmax': 0.2}, 'as': 'z'},
 'x1': {'func': <function SNeIaStretch.nicolas2021 at 0x2a1033640>},
 'c': {'func': <function SNeIaColor.color_rvs at 0x2a10329e0>},
 't0': {'func': <built-in method uniform of numpy.random.mtrand.RandomState object at 0x1100abe40>,
        'kwargs': {'low': 56000, 'high': 56200}},
 'magabs': {'func': <function SNeIaMagnitude.tripp1998 at 0x2a10336d0>,
            'kwargs': {'x1': '@x1',
                       'c': '@c',
                       'mabs': -19.3,
                       'sigmaint': 0.1,
                       'alpha': -0.15,
                       'beta': 3.8}},
 'magobs': {'func': 'magabs_to_magobs',
            'kwargs': {'z': '@z', 'magabs': '@magabs'}},
 'x0': {'func': 'magobs_to_amplitude',
        'kwargs': {'magobs': '@magobs', 'param_name': 'x0'}},
 'radec': {'func': <function random_radec at 0x17b8f0dc0>,
           'kwargs': {},
           'as': ['ra', 'dec']},
 'mwebv': {'func': <function get_mwebv at 0x2a1032d40>,
           'kwargs': {'ra': '@ra', 'dec': '@dec'}}}

---

Changing model entry’s (func and kwargs)

You can also change the way a parameter is drawn, not just it’s function parameters.

To do so, you need to update func and provide its new kwargs (if any). This means, updating the model for the entry.

To do that, you can use:

• the model option in draw(): this will update the model used to draw but will not affect the actual instance’s model

• the model option in from_draw(): this changes the model while creating the instance, it is thus permanent

• the update_model() method: this is permanent (it is what from_draw() uses).

The format is that of a model (see modelDAG):

{entry_name: {"func": a_function, ["kwargs": dict, "as": string_or_list]}}

For instance, say we want to change how the c parameter is drawn, we want to use a uniform distribution between -1 and +3:

import numpy as np
new_model_for_c = {"c": {"func": np.random.uniform, "kwargs": {"low":-1, "high":+3}}}

snia = skysurvey.SNeIa.from_draw(size=1_000, 
                                 model=new_model_for_c
                                )

ax = snia.data["c"].plot.hist()

here, the change is permanent, as you changed the model used to load the instance:

snia.model # check-out the c-entry

{'redshift': {'kwargs': {'zmax': 0.2}, 'as': 'z'},
 'x1': {'func': <function SNeIaStretch.nicolas2021 at 0x2a1033640>},
 'c': {'func': <built-in method uniform of numpy.random.mtrand.RandomState object at 0x1100abe40>,
       'kwargs': {'low': -1, 'high': 3}},
 't0': {'func': <built-in method uniform of numpy.random.mtrand.RandomState object at 0x1100abe40>,
        'kwargs': {'low': 56000, 'high': 56200}},
 'magabs': {'func': <function SNeIaMagnitude.tripp1998 at 0x2a10336d0>,
            'kwargs': {'x1': '@x1',
                       'c': '@c',
                       'mabs': -19.3,
                       'sigmaint': 0.1,
                       'alpha': -0.15,
                       'beta': 3.8}},
 'magobs': {'func': 'magabs_to_magobs',
            'kwargs': {'z': '@z', 'magabs': '@magabs'}},
 'x0': {'func': 'magobs_to_amplitude',
        'kwargs': {'magobs': '@magobs', 'param_name': 'x0'}},
 'radec': {'func': <function random_radec at 0x17b8f0dc0>,
           'kwargs': {},
           'as': ['ra', 'dec']},
 'mwebv': {'func': <function get_mwebv at 0x2a1032d40>,
           'kwargs': {'ra': '@ra', 'dec': '@dec'}}}

to change the model only for a draw() call

snia = skysurvey.SNeIa()
data_updated = snia.draw(size=1_000, model=new_model_for_c)
data_not_updated = snia.draw(size=1_000)

data_updated["c"].plot.hist()

<Axes: ylabel='Frequency'>

data_not_updated["c"].plot.hist()

<Axes: ylabel='Frequency'>

Use update_model()

You can directly change (permanently) the model using update_model:

flat_model = {"c": {"func": np.random.uniform, "kwargs": {"low":-0.1, "high":+1}},
              "x1": {"func": np.random.uniform, "kwargs": {"low":-3, "high":+3}}
             }

snia = skysurvey.SNeIa()
snia.update_model(**flat_model)
data = snia.draw(1_000)

data

	z	x1	c	t0	ra	dec	magabs	mwebv	magobs	x0
0	0.19865	1.176186	0.444875	56008.340695	222.896552	-1.865063	-17.675075	0.080541	22.335060	0.000018
1	0.09855	0.709318	0.437820	56096.020402	158.830749	-9.360366	-17.768205	0.038432	20.585214	0.000093
2	0.07125	-0.583655	-0.001735	56161.352299	40.159907	-24.963720	-19.342852	0.030212	18.266083	0.000784
3	0.06465	2.065721	0.245723	56099.688732	151.643297	62.479109	-18.483481	0.025641	18.904432	0.000435
4	0.18575	-2.142388	0.133901	56132.462905	160.724741	44.962650	-18.311956	0.024015	21.536072	0.000039
...	...	...	...	...	...	...	...	...	...	...
995	0.13945	0.925297	0.113981	56051.033345	101.707529	62.187247	-19.117665	0.152948	20.046828	0.000152
996	0.08255	0.951880	0.651901	56007.822847	334.024022	58.718666	-16.954604	1.691119	20.990792	0.000064
997	0.16525	2.239961	0.897326	56053.932034	191.704559	-51.507224	-16.329488	0.340121	23.238086	0.000008
998	0.15115	1.332900	0.508084	56145.351300	274.529018	36.913573	-17.509565	0.043677	21.845660	0.000029
999	0.12975	0.026171	0.742875	56032.272295	145.614963	-20.215206	-16.616113	0.074815	22.378542	0.000018

1000 rows × 10 columns

Shall I update a model or create a new one ?

model option can affect any entry, you can even add entry, but any new entry will be appended at the end of the dist model. So to create an internal dependency, you will need to create a new model from scratch.

---

How to insert a new entry to create an internal dependency

Say you want to randomly draw beta from a normal distribution and use this drawn parameter as input of magabs.

Using update_model() will work because modeldag orders the model entry prior to call draw(). Hence, beta will be inserted prior magabs the used model dict. However, do not forget to update the magabs kwargs to specify that beta must use the drawn beta.

random_beta_model = {"beta": {"func": np.random.normal, "kwargs": {"loc":3.5, "scale":0.1}}}

snia = skysurvey.SNeIa.from_draw(1_000, model=random_beta_model,
                                magabs={"beta":"@beta"}  # don't forget to update magabs
                                )

snia.model.visualize()

snia.data

	z	x1	c	t0	ra	dec	beta	magabs	mwebv	magobs	x0	template
0	0.18755	0.310	0.100449	56100.199219	91.745285	26.643330	3.502293	-19.082514	0.438025	20.788748	0.000077	salt2
1	0.12435	-0.080	0.106237	56125.371094	69.915680	21.941198	3.520313	-18.815409	0.282333	20.079466	0.000148	salt2
2	0.18605	-1.530	0.105854	56039.820312	239.642227	-24.382992	3.617054	-18.552702	0.372540	21.299213	0.000048	salt2
3	0.11905	-2.500	-0.031302	56158.765625	332.705963	-19.575117	3.390035	-19.192381	0.031475	19.600519	0.000229	salt2
4	0.12445	-1.280	-0.134878	56014.058594	124.406120	3.857391	3.389037	-19.595694	0.024401	19.301065	0.000302	salt2
...	...	...	...	...	...	...	...	...	...	...	...	...
995	0.17325	1.370	0.315792	56052.222656	1.105094	14.998715	3.482856	-18.437946	0.067036	21.242723	0.000051	salt2
996	0.17845	1.805	0.105748	56099.261719	297.663971	-42.456161	3.445384	-19.097910	0.086934	20.653700	0.000087	salt2
997	0.18925	0.125	-0.025275	56126.058594	289.028870	-4.340125	3.559816	-19.504374	0.641207	20.388643	0.000111	salt2
998	0.16945	-0.550	0.111878	56142.671875	3.465228	54.717876	3.474002	-18.913858	0.306152	20.713709	0.000082	salt2
999	0.12105	1.125	0.490803	56145.421875	75.796028	-32.925766	3.558682	-17.743340	0.023616	21.088533	0.000058	salt2

1000 rows × 12 columns