When selecting in time and displaying with DAS2 we seem to see a discontinuity in beteween time selections.
Need to check how tome range selections returns time and of there is no overlap / gap.

They should not be multiplied by 2.

Find a way to flagg the data on the dynamic spectrum level without filtering out astrophysical signal!

For some methods such as self.tmean(), np.datetime64 are used. Replace them according to the possible inputs and in any case with an astropy.time.Time object.

Override an opperator (maybe __or__(), as __and__() is for frequency concatenation) to allow for time concatenation.

Complete all docstrings

On the time and frequency averaged data (L1), correct for the known gain jumps at pointing switches and keep this information somewhere (it is likely to be stored within the ouptut FITS files).

Complete all docstrings

Add a method in Spectrum to average in time and frequency a dynamic spectrum selection.

For the frequency, modify directly within specdata.py and create a new method, something like self.frebin(...)...

Complete all docstrings

Cedric's code: https://gitlab.obspm.fr/cviou/nenufar_raw_tf_readers/-/blob/master/nenufar_raw.py

Enable the possibility of kwargs inputs in display methods in order to control the plotting output.
For vmin, vmax and cmap maybe do somethign like:

if 'vmin' not in kwargs.keys():
    kwargs['vmin'] = np.percentile(specdata.amp, 5)

...

After first analysis steps (averaging, flagging, data selction on a small sample, correction of pointing switch gain jumps...) depending on the NenuFAR key project,
Create a FITS file with:

• L1 data (n_time * n_freq * n_beam * n_stokes)

• Time (n_time)

• Freq (n_freq)

• Beam (n_beam)

• Stokes (n_stokes)

• Weight table (n_time * n_freq * n_beam * n_stokes): containing 0-1 values representing how much data was flagged to get the averaged point.

• pointing_switch_correction (n_time): table of correction to 'flatten' the data in time to compensate for the gain jumps due to analog pointing of the Mini-Arrays

• metadata : this is still a question, do we rely on BST files? Do we copy the entire BST file headers and HDUs (other than data)? Do we select only the relevant informations...?

Lane objects contains an error in _get_freq() which causes a shift in returned frequency, altough data is correct...

Simulate dynamic spectra with different signals (such as FRBs) to test for analysis algorithm on.

Apply a smoothing over the medianed time and frequency to get a smoother backgorund estimation.

Enable addition, subtraction, multiplication, division

Bug for this case:

from nenupytf.read import Spectrum
s = Spectrum('/databf2/nenufar-tf/20190320_104900_20190320_125000_SUN_TRACKING_BHR')
spec = s.average(time=['2019-03-20T10:49:41.0000000', '2019-03-20T10:49:42.0000000'], freq=[30, 35], dt=0.01)

Use .reshape array before multiplying by the bandpass rather than using np.tile() on the bandpass. This should accelerate the process.

To test :

from nenupytf.read import Spectrum
s = Spectrum('/databf2/nenufar-tf/2019/11/20191122_094700_20191122_130000_SUN_TRACKING')
%prun spec = s.select(time=['2019-11-22T09:47:55.000000', '2019-11-22T09:48:10.000000'], beam=0)

        236338 function calls (236248 primitive calls) in 1.288 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
        1    0.464    0.464    1.259    1.259 lane.py:297(select)
        1    0.324    0.324    0.554    0.554 stokes.py:92(correct)
       97    0.229    0.002    0.229    0.002 {method 'reduce' of 'numpy.ufunc' objects}

Bug for e.g.:

from nenupytf.read import Spectrum
s = Spectrum('/databf2/nenufar-tf/20190320_104900_20190320_125000_SUN_TRACKING_BHR')
spec = s.average(time=['2019-03-20T10:49:41.0000000', '2019-03-20T10:49:42.0000000'], freq=[30, 35], dt=0.0001)

from nenupytf.read import Spectrum
from nenupytf.display import plotdb
s = Spectrum('')
spec = s.average(time=['2019-11-22T09:59:30.0000000', '2019-11-22T10:04:30.0000000'], df=0.1, dt=0.1, stokes='I')
plotdb(spec)

Compared to

import matplotlib.pyplot as plt
plt.imshow(spec.db.T, origin='lower', aspect='auto')
plt.show()

Once the main functionnalities are working, edit the readme, making sure a standard user can find his way to use the package.