Add gaussian noise to a Target

You may want to generate a target with noise on the parameters without having to run the whole dataset + fitting process.

The apply_gaussian_noise method of the Target object is made for that.

Noise model follows the modeldag approach.

The errors are drawn using the modeldag structure: {"name": {"func": , "kwargs":{}}, }

This model specifies how the parameter error is drawn. Then, the “observed” parameter will be drawn from a normal distribution centered on the “true” parameter with the parameter error as width, so: param_obs = np.normal.rvs(loc=param_true, scale=param_err)

---

Most simple case, random noise on 1 parameter

Step 0: create a timeserie target based on the “v19-2005bf” sncosmo template.

import skysurvey
import numpy as np

target = skysurvey.TSTransient(template="v19-2005bf")
# specify the absolute magnitude of the target (nan by default)
target.set_magabs([-18, 1])

# draw 1000 target, use inplace=True to change self.data
_ = target.draw(1_000, inplace=True)
target.data.head(3)

	z	t0	magabs	ra	dec	magobs	amplitude	template
0	0.04635	56080.234375	-18.542274	217.788696	80.624512	18.094978	3.164150e-15	v19-2005bf
1	0.04025	56033.644531	-18.581287	317.605774	7.081192	17.740023	4.387723e-15	v19-2005bf
2	0.04865	56147.906250	-14.416636	145.240326	-56.762920	22.329351	6.404838e-17	v19-2005bf

Step 1: create an errormodel for magobs

Let’s draw error on magobs assuming they are \(0.1 ± 0.01\):

errormodel = {"magobs": {"func": np.random.normal, "kwargs":{"loc":0.1, "scale":0.01}} }

Step 2: apply gaussian noise to the target’s parameter

You will have a warning message, because the code doesn’t know who to draw the error.

noisy_target = target.apply_gaussian_noise(errormodel)
noisy_target.data.head(1)

	z	t0	magabs	ra	dec	magobs_true	amplitude	template	magobs_err	magobs
0	0.04635	56080.234375	-18.542274	217.788696	80.624512	18.094978	3.164150e-15	v19-2005bf	0.110416	18.045027

What happened ?

For each parameter given in the errormodel (here, only magobs):

1. rename {parameter} as {parameter}_true

2. draw {parameter}_{err} following the errormodel

3. draw {parameter} from a normal distribution centered on {parameter}_true with a width of {parameter}_{err}