def test_crosssubtract_datamodel(self): self.actualSetUp(zerow=True) future_vis = arlexecute.scatter(self.vis_list[0]) future_skymodel_list = arlexecute.scatter(self.skymodel_list) skymodel_vislist = predict_skymodel_list_arlexecute_workflow( future_vis, future_skymodel_list, context='2d', docal=True) skymodel_vislist = arlexecute.compute(skymodel_vislist, sync=True) vobs = sum_predict_results(skymodel_vislist) future_vobs = arlexecute.scatter(vobs) skymodel_vislist = crosssubtract_datamodels_skymodel_list_arlexecute_workflow( future_vobs, skymodel_vislist) skymodel_vislist = arlexecute.compute(skymodel_vislist, sync=True) result_skymodel = [ SkyModel(components=None, image=self.skymodel_list[-1].image) for v in skymodel_vislist ] self.vis_list = arlexecute.scatter(self.vis_list) result_skymodel = invert_skymodel_list_arlexecute_workflow( skymodel_vislist, result_skymodel, context='2d', docal=True) results = arlexecute.compute(result_skymodel, sync=True) assert numpy.max(numpy.abs(results[0][0].data)) > 0.0 assert numpy.max(numpy.abs(results[0][1])) > 0.0 if self.plot: import matplotlib.pyplot as plt from wrappers.arlexecute.image.operations import show_image show_image(results[0][0], title='Dirty image after cross-subtraction', vmax=0.1, vmin=-0.01) plt.show()
def test_predict(self): self.actualSetUp(zerow=True) self.skymodel_list = [ arlexecute.execute(create_low_test_skymodel_from_gleam)( npixel=self.npixel, cellsize=self.cellsize, frequency=[self.frequency[f]], phasecentre=self.phasecentre, polarisation_frame=PolarisationFrame("stokesI"), flux_limit=0.3, flux_threshold=0.3, flux_max=5.0) for f, freq in enumerate(self.frequency) ] self.skymodel_list = arlexecute.compute(self.skymodel_list, sync=True) assert isinstance(self.skymodel_list[0].images[0], Image), self.skymodel_list[0].images[0] assert isinstance(self.skymodel_list[0].components[0], Skycomponent), self.skymodel_list[0].components[0] assert len(self.skymodel_list[0].components) == 119, len( self.skymodel_list[0].components) assert len(self.skymodel_list[0].images) == 1, len( self.skymodel_list[0].images) assert numpy.max(numpy.abs( self.skymodel_list[0].images[0].data)) > 0.0, "Image is empty" self.skymodel_list = arlexecute.scatter(self.skymodel_list) skymodel_vislist = predict_skymodel_list_arlexecute_workflow( self.vis_list, self.skymodel_list, context='2d') skymodel_vislist = arlexecute.compute(skymodel_vislist, sync=True) assert numpy.max(numpy.abs(skymodel_vislist[0].vis)) > 0.0
def test_predictcal(self): self.actualSetUp(zerow=True) future_vis = arlexecute.scatter(self.vis_list[0]) future_skymodel = arlexecute.scatter(self.skymodel_list) skymodel_vislist = predict_skymodel_list_arlexecute_workflow(future_vis, future_skymodel, context='2d', docal=True) skymodel_vislist = arlexecute.compute(skymodel_vislist, sync=True) vobs = sum_predict_results(skymodel_vislist) if self.plot: def plotvis(i, v): import matplotlib.pyplot as plt uvr = numpy.hypot(v.u, v.v) amp = numpy.abs(v.vis[:, 0]) plt.plot(uvr, amp, '.') plt.title(str(i)) plt.show() plotvis(0, vobs)
def actualSetup(self, nsources=None, nvoronoi=None): n_workers = 8 # Set up the observation: 10 minutes at transit, with 10s integration. # Skip 5/6 points to avoid outstation redundancy nfreqwin = 1 ntimes = 3 self.rmax = 2500.0 dec = -40.0 * u.deg frequency = [1e8] channel_bandwidth = [0.1e8] times = numpy.linspace(-10.0, 10.0, ntimes) * numpy.pi / (3600.0 * 12.0) phasecentre = SkyCoord(ra=+0.0 * u.deg, dec=dec, frame='icrs', equinox='J2000') low = create_named_configuration('LOWBD2', rmax=self.rmax) centre = numpy.mean(low.xyz, axis=0) distance = numpy.hypot(low.xyz[:, 0] - centre[0], low.xyz[:, 1] - centre[1], low.xyz[:, 2] - centre[2]) lowouter = low.data[distance > 1000.0][::6] lowcore = low.data[distance < 1000.0][::3] low.data = numpy.hstack((lowcore, lowouter)) blockvis = create_blockvisibility( low, times, frequency=frequency, channel_bandwidth=channel_bandwidth, weight=1.0, phasecentre=phasecentre, polarisation_frame=PolarisationFrame("stokesI"), zerow=True) vis = convert_blockvisibility_to_visibility(blockvis) advice = advise_wide_field(vis, guard_band_image=2.0, delA=0.02) cellsize = advice['cellsize'] npixel = advice['npixels2'] small_model = create_image_from_visibility(blockvis, npixel=512, frequency=frequency, nchan=nfreqwin, cellsize=cellsize, phasecentre=phasecentre) vis.data['imaging_weight'][...] = vis.data['weight'][...] vis = weight_list_serial_workflow([vis], [small_model])[0] vis = taper_list_serial_workflow([vis], 3 * cellsize)[0] blockvis = convert_visibility_to_blockvisibility(vis) # ### Generate the model from the GLEAM catalog, including application of the primary beam. beam = create_image_from_visibility(blockvis, npixel=npixel, frequency=frequency, nchan=nfreqwin, cellsize=cellsize, phasecentre=phasecentre) beam = create_low_test_beam(beam, use_local=False) flux_limit = 0.5 original_gleam_components = create_low_test_skycomponents_from_gleam( flux_limit=flux_limit, phasecentre=phasecentre, frequency=frequency, polarisation_frame=PolarisationFrame('stokesI'), radius=0.15) all_components = apply_beam_to_skycomponent(original_gleam_components, beam) all_components = filter_skycomponents_by_flux(all_components, flux_min=flux_limit) voronoi_components = filter_skycomponents_by_flux(all_components, flux_min=1.5) def max_flux(elem): return numpy.max(elem.flux) voronoi_components = sorted(voronoi_components, key=max_flux, reverse=True) if nsources is not None: all_components = [all_components[0]] if nvoronoi is not None: voronoi_components = [voronoi_components[0]] self.screen = import_image_from_fits( arl_path('data/models/test_mpc_screen.fits')) all_gaintables = create_gaintable_from_screen(blockvis, all_components, self.screen) gleam_skymodel_noniso = [ SkyModel(components=[all_components[i]], gaintable=all_gaintables[i]) for i, sm in enumerate(all_components) ] # ### Now predict the visibility for each skymodel and apply the gaintable for that skymodel, # returning a list of visibilities, one for each skymodel. We then sum these to obtain # the total predicted visibility. All images and skycomponents in the same skymodel # get the same gaintable applied which means that in this case each skycomponent has a separate gaintable. self.all_skymodel_noniso_vis = convert_blockvisibility_to_visibility( blockvis) ngroup = n_workers future_vis = arlexecute.scatter(self.all_skymodel_noniso_vis) chunks = [ gleam_skymodel_noniso[i:i + ngroup] for i in range(0, len(gleam_skymodel_noniso), ngroup) ] for chunk in chunks: result = predict_skymodel_list_arlexecute_workflow(future_vis, chunk, context='2d', docal=True) work_vis = arlexecute.compute(result, sync=True) for w in work_vis: self.all_skymodel_noniso_vis.data['vis'] += w.data['vis'] assert numpy.max( numpy.abs(self.all_skymodel_noniso_vis.data['vis'])) > 0.0 self.all_skymodel_noniso_blockvis = convert_visibility_to_blockvisibility( self.all_skymodel_noniso_vis) # ### Remove weaker of components that are too close (0.02 rad) idx, voronoi_components = remove_neighbouring_components( voronoi_components, 0.02) model = create_image_from_visibility(blockvis, npixel=npixel, frequency=frequency, nchan=nfreqwin, cellsize=cellsize, phasecentre=phasecentre) # Use the gaintable for the brightest component as the starting gaintable all_gaintables[0].gain[...] = numpy.conjugate( all_gaintables[0].gain[...]) all_gaintables[0].gain[...] = 1.0 + 0.0j self.theta_list = initialize_skymodel_voronoi(model, voronoi_components, all_gaintables[0])
def trial_case(results, seed=180555, context='wstack', nworkers=8, threads_per_worker=1, memory=8, processes=True, order='frequency', nfreqwin=7, ntimes=3, rmax=750.0, facets=1, wprojection_planes=1, use_dask=True, use_serial_imaging=True, flux_limit=0.3, nmajor=5, dft_threshold=1.0, use_serial_clean=True, write_fits=False): """ Single trial for performance-timings Simulates visibilities from GLEAM including phase errors Makes dirty image and PSF Runs ICAL pipeline The results are in a dictionary: 'context': input - a string describing concisely the purpose of the test 'time overall', overall execution time (s) 'time predict', time to execute GLEAM prediction graph 'time invert', time to make dirty image 'time invert graph', time to make dirty image graph 'time ICAL graph', time to create ICAL graph 'time ICAL', time to execute ICAL graph 'context', type of imaging e.g. 'wstack' 'nworkers', number of workers to create 'threads_per_worker', 'nnodes', Number of nodes, 'processes', 'order', Ordering of data_models 'nfreqwin', Number of frequency windows in simulation 'ntimes', Number of hour angles in simulation 'rmax', Maximum radius of stations used in simulation (m) 'facets', Number of facets in deconvolution and imaging 'wprojection_planes', Number of wprojection planes 'vis_slices', Number of visibility slices (per Visibbility) 'npixel', Number of pixels in image 'cellsize', Cellsize in radians 'seed', Random number seed 'dirty_max', Maximum in dirty image 'dirty_min', Minimum in dirty image 'restored_max', 'restored_min', 'deconvolved_max', 'deconvolved_min', 'residual_max', 'residual_min', 'git_info', GIT hash (not definitive since local mods are possible) :param results: Initial state :param seed: Random number seed (used in gain simulations) :param context: imaging context :param context: Type of context: '2d'|'timeslice'|'wstack' :param nworkers: Number of dask workers to use :param threads_per_worker: Number of threads per worker :param processes: Use processes instead of threads 'processes'|'threads' :param order: See simulate_list_list_arlexecute_workflow_workflowkflow :param nfreqwin: See simulate_list_list_arlexecute_workflow_workflowkflow :param ntimes: See simulate_list_list_arlexecute_workflow_workflowkflow :param rmax: See simulate_list_list_arlexecute_workflow_workflowkflow :param facets: Number of facets to use :param wprojection_planes: Number of wprojection planes to use :param use_dask: Use dask or immediate evaluation :return: results dictionary """ if use_dask: client = get_dask_Client(threads_per_worker=threads_per_worker, processes=threads_per_worker == 1, memory_limit=memory * 1024 * 1024 * 1024, n_workers=nworkers) arlexecute.set_client(client) nodes = findNodes(arlexecute.client) print("Defined %d workers on %d nodes" % (nworkers, len(nodes))) print("Workers are: %s" % str(nodes)) else: arlexecute.set_client(use_dask=use_dask) results['nnodes'] = 1 def init_logging(): logging.basicConfig( filename='pipelines_arlexecute_timings.log', filemode='w', format='%(asctime)s,%(msecs)d %(name)s %(levelname)s %(message)s', datefmt='%H:%M:%S', level=logging.INFO) init_logging() log = logging.getLogger() # Initialise logging on the workers. This appears to only work using the process scheduler. arlexecute.run(init_logging) def lprint(*args): log.info(*args) print(*args) lprint("Starting pipelines_arlexecute_timings") numpy.random.seed(seed) results['seed'] = seed start_all = time.time() results['context'] = context results['hostname'] = socket.gethostname() results['git_hash'] = git_hash() results['epoch'] = time.strftime("%Y-%m-%d %H:%M:%S") lprint("Context is %s" % context) results['nworkers'] = nworkers results['threads_per_worker'] = threads_per_worker results['processes'] = processes results['memory'] = memory results['order'] = order results['nfreqwin'] = nfreqwin results['ntimes'] = ntimes results['rmax'] = rmax results['facets'] = facets results['wprojection_planes'] = wprojection_planes results['dft threshold'] = dft_threshold results['use_dask'] = use_dask lprint("At start, configuration is:") lprint(sort_dict(results)) # Parameters determining scale of simulation. frequency = numpy.linspace(1.0e8, 1.2e8, nfreqwin) centre = nfreqwin // 2 if nfreqwin > 1: channel_bandwidth = numpy.array(nfreqwin * [frequency[1] - frequency[0]]) else: channel_bandwidth = numpy.array([1e6]) times = numpy.linspace(-numpy.pi / 4.0, numpy.pi / 4.0, ntimes) phasecentre = SkyCoord(ra=+0.0 * u.deg, dec=-40.0 * u.deg, frame='icrs', equinox='J2000') lprint("****** Visibility creation ******") # Create the empty BlockVisibility's and persist these on the cluster tmp_bvis_list = simulate_list_arlexecute_workflow( 'LOWBD2', frequency=frequency, channel_bandwidth=channel_bandwidth, times=times, phasecentre=phasecentre, order=order, format='blockvis', rmax=rmax) tmp_vis_list = [ arlexecute.execute(convert_blockvisibility_to_visibility)(bv) for bv in tmp_bvis_list ] tmp_vis_list = arlexecute.client.compute(tmp_vis_list, sync=True) vis_list = arlexecute.gather(tmp_vis_list) import matplotlib.pyplot as plt plt.clf() plt.hist(vis_list[0].w, bins=100) plt.title('Histogram of w samples: rms=%.1f (wavelengths)' % numpy.std(vis_list[0].w)) plt.xlabel('W (wavelengths)') plt.show() plt.clf() plt.hist(vis_list[0].uvdist, bins=100) plt.title('Histogram of uvdistance samples') plt.xlabel('UV Distance (wavelengths)') plt.show() arlexecute.client.cancel(tmp_vis_list) future_vis_list = arlexecute.scatter(vis_list) # Find the best imaging parameters but don't bring the vis_list back here print("****** Finding wide field parameters ******") future_advice = [ arlexecute.execute(advise_wide_field)( v, guard_band_image=6.0, delA=0.1, facets=facets, wprojection_planes=wprojection_planes, oversampling_synthesised_beam=4.0) for v in future_vis_list ] future_advice = arlexecute.compute(future_advice) advice = arlexecute.client.gather(future_advice)[-1] arlexecute.client.cancel(future_advice) # Deconvolution via sub-images requires 2^n npixel = advice['npixels2'] results['npixel'] = npixel cellsize = advice['cellsize'] results['cellsize'] = cellsize lprint("Image will have %d by %d pixels, cellsize = %.6f rad" % (npixel, npixel, cellsize)) # Create an empty model image tmp_model_list = [ arlexecute.execute(create_image)( npixel=npixel, cellsize=cellsize, frequency=[frequency[f]], channel_bandwidth=[channel_bandwidth[f]], phasecentre=phasecentre, polarisation_frame=PolarisationFrame("stokesI")) for f, freq in enumerate(frequency) ] model_list = arlexecute.compute(tmp_model_list, sync=True) future_model_list = arlexecute.scatter(model_list) lprint("****** Setting up imaging parameters ******") # Now set up the imaging parameters template_model = create_image( npixel=npixel, cellsize=cellsize, frequency=[frequency[centre]], phasecentre=phasecentre, channel_bandwidth=[channel_bandwidth[centre]], polarisation_frame=PolarisationFrame("stokesI")) gcfcf = [create_pswf_convolutionfunction(template_model)] if context == 'timeslice': vis_slices = ntimes lprint("Using timeslice with %d slices" % vis_slices) elif context == '2d': vis_slices = 1 elif context == "wprojection": wstep = advice['wstep'] nw = advice['wprojection_planes'] vis_slices = 1 support = advice['nwpixels'] results['wprojection_planes'] = nw lprint("****** Starting W projection kernel creation ******") lprint("Using wprojection with %d planes with wstep %.1f wavelengths" % (nw, wstep)) lprint("Support of wprojection = %d pixels" % support) gcfcf = [ create_awterm_convolutionfunction(template_model, nw=nw, wstep=wstep, oversampling=4, support=support, use_aaf=True) ] lprint("Size of W projection gcf, cf = %.2E bytes" % get_size(gcfcf)) else: context = 'wstack' vis_slices = advice['vis_slices'] lprint("Using wstack with %d slices" % vis_slices) gcfcf = arlexecute.scatter(gcfcf, broadcast=True) results['vis_slices'] = vis_slices # Make a skymodel from gleam, with bright sources as components and weak sources in an image lprint("****** Starting GLEAM skymodel creation ******") future_skymodel_list = [ arlexecute.execute(create_low_test_skymodel_from_gleam)( npixel=npixel, cellsize=cellsize, frequency=[frequency[f]], phasecentre=phasecentre, polarisation_frame=PolarisationFrame("stokesI"), flux_limit=flux_limit, flux_threshold=dft_threshold, flux_max=5.0) for f, freq in enumerate(frequency) ] # We use predict_skymodel so that we can use skycomponents as well as images lprint("****** Starting GLEAM skymodel prediction ******") predicted_vis_list = [ predict_skymodel_list_arlexecute_workflow(future_vis_list[f], [future_skymodel_list[f]], context=context, vis_slices=vis_slices, facets=facets, gcfcf=gcfcf)[0] for f, freq in enumerate(frequency) ] # Corrupt the visibility for the GLEAM model lprint("****** Visibility corruption ******") tmp_corrupted_vis_list = corrupt_list_arlexecute_workflow( predicted_vis_list, phase_error=1.0, seed=seed) lprint("****** Weighting and tapering ******") tmp_corrupted_vis_list = weight_list_arlexecute_workflow( tmp_corrupted_vis_list, future_model_list) tmp_corrupted_vis_list = taper_list_arlexecute_workflow( tmp_corrupted_vis_list, 0.003 * 750.0 / rmax) tmp_corrupted_vis_list = arlexecute.compute(tmp_corrupted_vis_list, sync=True) corrupted_vis_list = arlexecute.gather(tmp_corrupted_vis_list) # arlexecute.client.cancel(predicted_vis_list) arlexecute.client.cancel(tmp_corrupted_vis_list) future_corrupted_vis_list = arlexecute.scatter(corrupted_vis_list) # At this point the only futures are of scatter'ed data so no repeated calculations should be # incurred. lprint("****** Starting dirty image calculation ******") start = time.time() dirty_list = invert_list_arlexecute_workflow( future_corrupted_vis_list, future_model_list, vis_slices=vis_slices, context=context, facets=facets, use_serial_invert=use_serial_imaging, gcfcf=gcfcf) results['size invert graph'] = get_size(dirty_list) lprint('Size of dirty graph is %.3E bytes' % (results['size invert graph'])) end = time.time() results['time invert graph'] = end - start lprint("Construction of invert graph took %.3f seconds" % (end - start)) start = time.time() dirty, sumwt = arlexecute.compute(dirty_list, sync=True)[centre] end = time.time() results['time invert'] = end - start lprint("Dirty image invert took %.3f seconds" % (end - start)) lprint("Maximum in dirty image is %f, sumwt is %s" % (numpy.max(numpy.abs(dirty.data)), str(sumwt))) qa = qa_image(dirty) results['dirty_max'] = qa.data['max'] results['dirty_min'] = qa.data['min'] if write_fits: export_image_to_fits( dirty, "pipelines_arlexecute_timings-%s-dirty.fits" % context) lprint("****** Starting prediction ******") start = time.time() tmp_vis_list = predict_list_arlexecute_workflow( future_corrupted_vis_list, future_model_list, vis_slices=vis_slices, context=context, facets=facets, use_serial_predict=use_serial_imaging, gcfcf=gcfcf) result = arlexecute.compute(tmp_vis_list, sync=True) # arlexecute.client.cancel(tmp_vis_list) end = time.time() results['time predict'] = end - start lprint("Predict took %.3f seconds" % (end - start)) # Create the ICAL pipeline to run major cycles, starting selfcal at cycle 1. A global solution across all # frequencies (i.e. Visibilities) is performed. print("Using subimage clean") deconvolve_facets = 8 deconvolve_overlap = 16 deconvolve_taper = 'tukey' lprint("****** Starting ICAL graph creation ******") controls = create_calibration_controls() controls['T']['first_selfcal'] = 1 controls['T']['timescale'] = 'auto' start = time.time() ical_list = ical_list_arlexecute_workflow( future_corrupted_vis_list, model_imagelist=future_model_list, context=context, vis_slices=vis_slices, scales=[0, 3, 10], algorithm='mmclean', nmoment=3, niter=1000, fractional_threshold=0.1, threshold=0.01, nmajor=nmajor, gain=0.25, psf_support=64, deconvolve_facets=deconvolve_facets, deconvolve_overlap=deconvolve_overlap, deconvolve_taper=deconvolve_taper, timeslice='auto', global_solution=True, do_selfcal=True, calibration_context='T', controls=controls, use_serial_predict=use_serial_imaging, use_serial_invert=use_serial_imaging, use_serial_clean=use_serial_clean, gcfcf=gcfcf) results['size ICAL graph'] = get_size(ical_list) lprint('Size of ICAL graph is %.3E bytes' % results['size ICAL graph']) end = time.time() results['time ICAL graph'] = end - start lprint("Construction of ICAL graph took %.3f seconds" % (end - start)) print("Current objects on cluster: ") pp.pprint(arlexecute.client.who_has()) # # Execute the graph lprint("****** Executing ICAL graph ******") start = time.time() deconvolved, residual, restored, gaintables = arlexecute.compute(ical_list, sync=True) end = time.time() results['time ICAL'] = end - start lprint("ICAL graph execution took %.3f seconds" % (end - start)) qa = qa_image(deconvolved[centre]) results['deconvolved_max'] = qa.data['max'] results['deconvolved_min'] = qa.data['min'] deconvolved_cube = image_gather_channels(deconvolved) if write_fits: export_image_to_fits( deconvolved_cube, "pipelines_arlexecute_timings-%s-ical_deconvolved.fits" % context) qa = qa_image(residual[centre][0]) results['residual_max'] = qa.data['max'] results['residual_min'] = qa.data['min'] residual_cube = remove_sumwt(residual) residual_cube = image_gather_channels(residual_cube) if write_fits: export_image_to_fits( residual_cube, "pipelines_arlexecute_timings-%s-ical_residual.fits" % context) qa = qa_image(restored[centre]) results['restored_max'] = qa.data['max'] results['restored_min'] = qa.data['min'] restored_cube = image_gather_channels(restored) if write_fits: export_image_to_fits( restored_cube, "pipelines_arlexecute_timings-%s-ical_restored.fits" % context) # arlexecute.close() end_all = time.time() results['time overall'] = end_all - start_all lprint("At end, results are:") results = sort_dict(results) lprint(results) return results
def mpccal_skymodel_list_arlexecute_workflow(visobs, model, theta_list, nmajor=10, context='2d', mpccal_progress=None, **kwargs): """Run MPC pipeline This runs the Model Partition Calibration algorithm. See SDP Memo 97 for more details, and see workflows/scripts/pipelines/mpccal_arlexecute_pipeline.py for an example of the application :param visobs: Visibility (not a list!) :param model: Model image :param theta_list: List of SkyModels i.e. theta in memo 97. :param nmajor: Number of major cycles :param context: Imaging context :param mpccal_progress: Function to display progress :return: Delayed tuple (theta_list, residual) """ psf_obs = invert_list_arlexecute_workflow([visobs], [model], context=context, dopsf=True) result = arlexecute.execute((theta_list, model)) for iteration in range(nmajor): # The E step of decoupling the data models vdatamodel_list = predict_skymodel_list_arlexecute_workflow( visobs, theta_list, context=context, docal=True, **kwargs) vdatamodel_list = crosssubtract_datamodels_skymodel_list_arlexecute_workflow( visobs, vdatamodel_list) # The M step: 1 - Update the models by deconvolving the residual image. The residual image must be calculated # from a difference of the dirty images from the data model, and the dirty images dirty_all_conv = convolve_skymodel_list_arlexecute_workflow( visobs, theta_list, context=context, docal=True, **kwargs) dirty_all_cal = invert_skymodel_list_arlexecute_workflow( vdatamodel_list, theta_list, context=context, docal=True, **kwargs) def diff_dirty(dcal, dconv): assert numpy.max(numpy.abs( dcal[0].data)) > 0.0, "before: dcal subimage is zero" dcal[0].data -= dconv[0].data assert numpy.max(numpy.abs( dcal[0].data)) > 0.0, "after: dcal subimage is zero" return dcal dirty_all_cal = [ arlexecute.execute(diff_dirty, nout=1)(dirty_all_cal[i], dirty_all_conv[i]) for i in range(len(dirty_all_cal)) ] def make_residual(dcal, tl, it): res = create_empty_image_like(dcal[0][0]) for i, d in enumerate(dcal): assert numpy.max(numpy.abs( d[0].data)) > 0.0, "Residual subimage is zero" if tl[i].mask is None: res.data += d[0].data else: assert numpy.max(numpy.abs( tl[i].mask.data)) > 0.0, "Mask image is zero" res.data += d[0].data * tl[i].mask.data assert numpy.max(numpy.abs( res.data)) > 0.0, "Residual image is zero" # import matplotlib.pyplot as plt # from processing_components.image.operations import show_image # show_image(res, title='MPCCAL residual image, iteration %d' % it) # plt.show() return res residual = arlexecute.execute(make_residual, nout=1)(dirty_all_cal, theta_list, iteration) deconvolved = deconvolve_list_arlexecute_workflow([(residual, 1.0)], [psf_obs[0]], [model], **kwargs) # The M step: 2 - Update the gaintables vpredicted_list = predict_skymodel_list_arlexecute_workflow( visobs, theta_list, context=context, docal=True, **kwargs) vcalibrated_list, gaintable_list = calibrate_list_arlexecute_workflow( vdatamodel_list, vpredicted_list, calibration_context='T', iteration=iteration, global_solution=False, **kwargs) if mpccal_progress is not None: theta_list = arlexecute.execute(mpccal_progress, nout=len(theta_list))( residual, theta_list, gaintable_list, iteration) theta_list = \ arlexecute.execute(update_skymodel_from_image, nout=len(theta_list))(theta_list, deconvolved[0]) theta_list = arlexecute.execute(update_skymodel_from_gaintables, nout=len(theta_list))( theta_list, gaintable_list, calibration_context='T') result = arlexecute.execute((theta_list, residual)) return result
####################################################################################################### # Calculate visibility by using the predict_skymodel function which applies a different gaintable table # for each skymodel. We do the calculation in chunks of nworkers skymodels. all_skymodel_blockvis = copy_visibility(block_vis, zero=True) all_skymodel_vis = convert_blockvisibility_to_visibility( all_skymodel_blockvis) ngroup = 8 future_vis = arlexecute.scatter(all_skymodel_vis) chunks = [ all_skymodel[i:i + ngroup] for i in range(0, len(all_skymodel), ngroup) ] for chunk in chunks: result = predict_skymodel_list_arlexecute_workflow(future_vis, chunk, context='2d', docal=True) work_vis = arlexecute.compute(result, sync=True) for w in work_vis: all_skymodel_vis.data['vis'] += w.data['vis'] assert numpy.max(numpy.abs(all_skymodel_vis.data['vis'])) > 0.0 all_skymodel_blockvis = convert_visibility_to_blockvisibility( all_skymodel_vis) ####################################################################################################### # Now proceed to run MPCCAL in ICAL mode i.e. with only one skymodel def progress(res, tl_list, gt_list, it, context='MPCCAL'): print('Iteration %d' % it) print(
def trial_case(results, seed=180555, context='wstack', nworkers=8, threads_per_worker=1, memory=8, processes=True, order='frequency', nfreqwin=7, ntimes=3, rmax=750.0, facets=1, wprojection_planes=1, use_dask=True, use_serial_imaging=False, flux_limit=0.3, nmajor=5, dft_threshold=1.0): """ Single trial for performance-timings Simulates visibilities from GLEAM including phase errors Makes dirty image and PSF Runs ICAL pipeline The results are in a dictionary: 'context': input - a string describing concisely the purpose of the test 'time overall', overall execution time (s) 'time create gleam', time to create GLEAM prediction graph 'time predict', time to execute GLEAM prediction graph 'time corrupt', time to corrupt data_models 'time invert', time to make dirty image 'time psf invert', time to make PSF 'time ICAL graph', time to create ICAL graph 'time ICAL', time to execute ICAL graph 'context', type of imaging e.g. 'wstack' 'nworkers', number of workers to create 'threads_per_worker', 'nnodes', Number of nodes, 'processes', 'order', Ordering of data_models 'nfreqwin', Number of frequency windows in simulation 'ntimes', Number of hour angles in simulation 'rmax', Maximum radius of stations used in simulation (m) 'facets', Number of facets in deconvolution and imaging 'wprojection_planes', Number of wprojection planes 'vis_slices', Number of visibility slices (per Visibbility) 'npixel', Number of pixels in image 'cellsize', Cellsize in radians 'seed', Random number seed 'dirty_max', Maximum in dirty image 'dirty_min', Minimum in dirty image 'psf_max', 'psf_min', 'restored_max', 'restored_min', 'deconvolved_max', 'deconvolved_min', 'residual_max', 'residual_min', 'git_info', GIT hash (not definitive since local mods are possible) :param results: Initial state :param seed: Random number seed (used in gain simulations) :param context: imaging context :param context: Type of context: '2d'|'timeslice'|'wstack' :param nworkers: Number of dask workers to use :param threads_per_worker: Number of threads per worker :param processes: Use processes instead of threads 'processes'|'threads' :param order: See simulate_list_list_arlexecute_workflow_workflowkflow :param nfreqwin: See simulate_list_list_arlexecute_workflow_workflowkflow :param ntimes: See simulate_list_list_arlexecute_workflow_workflowkflow :param rmax: See simulate_list_list_arlexecute_workflow_workflowkflow :param facets: Number of facets to use :param wprojection_planes: Number of wprojection planes to use :param use_dask: Use dask or immediate evaluation :return: results dictionary """ if use_dask: client = get_dask_Client(threads_per_worker=threads_per_worker, processes = threads_per_worker == 1, memory_limit=memory * 1024 * 1024 * 1024, n_workers=nworkers) arlexecute.set_client(client) nodes = findNodes(arlexecute.client) print("Defined %d workers on %d nodes" % (nworkers, len(nodes))) print("Workers are: %s" % str(nodes)) else: arlexecute.set_client(use_dask=use_dask) results['nnodes'] = 1 def init_logging(): logging.basicConfig(filename='pipelines-arlexecute-timings.log', filemode='a', format='%(asctime)s,%(msecs)d %(name)s %(levelname)s %(message)s', datefmt='%H:%M:%S', level=logging.INFO) init_logging() log = logging.getLogger() # Initialise logging on the workers. This appears to only work using the process scheduler. arlexecute.run(init_logging) def lprint(s): log.info(s) print(s) lprint("Starting pipelines-arlexecute-timings") numpy.random.seed(seed) results['seed'] = seed start_all = time.time() results['context'] = context results['hostname'] = socket.gethostname() results['git_hash'] = git_hash() results['epoch'] = time.strftime("%Y-%m-%d %H:%M:%S") lprint("Context is %s" % context) results['nworkers'] = nworkers results['threads_per_worker'] = threads_per_worker results['processes'] = processes results['memory'] = memory results['order'] = order results['nfreqwin'] = nfreqwin results['ntimes'] = ntimes results['rmax'] = rmax results['facets'] = facets results['wprojection_planes'] = wprojection_planes results['dft threshold'] = dft_threshold results['use_dask'] = use_dask lprint("At start, configuration is:") lprint(results) # Parameters determining scale frequency = numpy.linspace(1.0e8, 1.2e8, nfreqwin) centre = nfreqwin // 2 if nfreqwin > 1: channel_bandwidth = numpy.array(nfreqwin * [frequency[1] - frequency[0]]) else: channel_bandwidth = numpy.array([1e6]) times = numpy.linspace(-numpy.pi / 4.0, numpy.pi / 4.0, ntimes) phasecentre = SkyCoord(ra=+30.0 * u.deg, dec=-60.0 * u.deg, frame='icrs', equinox='J2000') bvis_list = simulate_list_arlexecute_workflow('LOWBD2', frequency=frequency, channel_bandwidth=channel_bandwidth, times=times, phasecentre=phasecentre, order=order, format='blockvis', rmax=rmax) lprint("****** Visibility creation ******") bvis_list = arlexecute.compute(bvis_list, sync=True) vis_list = [arlexecute.execute(convert_blockvisibility_to_visibility(bv)) for bv in bvis_list] vis_list = arlexecute.compute(vis_list, sync=True) # Find the best imaging parameters but don't bring the vis_list back here def get_wf(v): return advise_wide_field(v, guard_band_image=6.0, delA=0.1, facets=facets, wprojection_planes=wprojection_planes, oversampling_synthesised_beam=4.0) advice = arlexecute.compute(arlexecute.execute(get_wf)(vis_list[-1]), sync=True) # Deconvolution via sub-images requires 2^n npixel = advice['npixels2'] results['npixel'] = npixel cellsize = advice['cellsize'] results['cellsize'] = cellsize lprint("Image will have %d by %d pixels, cellsize = %.6f rad" % (npixel, npixel, cellsize)) # Create an empty model image model_list = [arlexecute.execute(create_image_from_visibility) (vis_list[f], npixel=npixel, cellsize=cellsize, frequency=[frequency[f]], channel_bandwidth=[channel_bandwidth[f]], polarisation_frame=PolarisationFrame("stokesI")) for f, freq in enumerate(frequency)] model_list = arlexecute.compute(model_list, sync=True) model_list = arlexecute.scatter(model_list) start = time.time() vis_list = weight_list_arlexecute_workflow(vis_list, model_list) vis_list = taper_list_arlexecute_workflow(vis_list, 0.003 * 750.0 / rmax) print("****** Starting weighting and tapering ******") vis_list = arlexecute.compute(vis_list, sync=True) end = time.time() results['time weight'] = end - start print("Weighting took %.3f seconds" % (end - start)) vis_list = arlexecute.scatter(vis_list) # Now set up the imaging parameters gcfcf_list = [None for i in range(nfreqwin)] if context == 'timeslice': vis_slices = ntimes lprint("Using timeslice with %d slices" % vis_slices) elif context == '2d': vis_slices = 1 elif context == "wprojection": wstep = advice['wstep'] nw = advice['wprojection_planes'] vis_slices = 1 support = advice['nwpixels'] results['wprojection_planes'] = nw lprint("Using wprojection with %d planes with wstep %.1f wavelengths" % (nw, wstep)) start = time.time() lprint("****** Starting W projection kernel creation ******") gcfcf_list = [arlexecute.execute(create_awterm_convolutionfunction, nout=1) (m, nw=nw, wstep=wstep, oversampling=8, support=support, use_aaf=True) for m in model_list] gcfcf_list = arlexecute.compute(gcfcf_list, sync=True) end = time.time() results['time create wprojection'] = end - start lprint("Creating W projection kernel took %.3f seconds" % (end - start)) cf_image = convert_convolutionfunction_to_image(gcfcf_list[centre][1]) cf_image.data = numpy.real(cf_image.data) export_image_to_fits(cf_image, "pipelines-arlexecute-timings-wterm-cf.fits") gcfcf_list = arlexecute.scatter(gcfcf_list) else: context = 'wstack' vis_slices = advice['vis_slices'] lprint("Using wstack with %d slices" % vis_slices) results['vis_slices'] = vis_slices # Make a skymodel from gleam, with bright sources as components and weak sources in an image lprint("****** Starting GLEAM skymodel creation ******") start = time.time() skymodel_list = [arlexecute.execute(create_low_test_skymodel_from_gleam) (npixel=npixel, cellsize=cellsize, frequency=[frequency[f]], phasecentre=phasecentre, polarisation_frame=PolarisationFrame("stokesI"), flux_limit=flux_limit, flux_threshold=dft_threshold, flux_max=5.0) for f, freq in enumerate(frequency)] skymodel_list = arlexecute.compute(skymodel_list, sync=True) end = time.time() lprint("GLEAM skymodel creation took %.3f seconds" % (end - start)) results['time create gleam'] = end - start lprint("****** Starting GLEAM skymodel prediction ******") start = time.time() predicted_vis_list = [predict_skymodel_list_arlexecute_workflow(vis_list[f], [skymodel_list[f]], context=context, vis_slices=vis_slices, facets=facets, gcfcf=[gcfcf_list[f]])[0] for f, freq in enumerate(frequency)] predicted_vis_list = arlexecute.compute(predicted_vis_list, sync=True) end = time.time() lprint("GLEAM skymodel prediction took %.3f seconds" % (end - start)) results['time predict gleam'] = end - start lprint("****** Starting psf image calculation ******") start = time.time() predicted_vis_list = arlexecute.scatter(predicted_vis_list) psf_list = invert_list_arlexecute_workflow(predicted_vis_list, model_list, vis_slices=vis_slices, dopsf=True, context=context, facets=facets, use_serial_invert=use_serial_imaging, gcfcf=gcfcf_list) psf, sumwt = arlexecute.compute(psf_list, sync=True)[centre] end = time.time() results['time psf invert'] = end - start lprint("PSF invert took %.3f seconds" % (end - start)) lprint("Maximum in psf image is %f, sumwt is %s" % (numpy.max(numpy.abs(psf.data)), str(sumwt))) qa = qa_image(psf) results['psf_max'] = qa.data['max'] results['psf_min'] = qa.data['min'] export_image_to_fits(psf, "pipelines-arlexecute-timings-%s-psf.fits" % context) # Make a smoothed model image for comparison # smoothed_model_list = restore_list_arlexecute_workflow(gleam_model_list, psf_list) # smoothed_model_list = arlexecute.compute(smoothed_model_list, sync=True) # smoothed_cube = image_gather_channels(smoothed_model_list) # export_image_to_fits(smoothed_cube, "pipelines-arlexecute-timings-cmodel.fits") # Create an empty model image model_list = [arlexecute.execute(create_image_from_visibility) (predicted_vis_list[f], npixel=npixel, cellsize=cellsize, frequency=[frequency[f]], channel_bandwidth=[channel_bandwidth[f]], polarisation_frame=PolarisationFrame("stokesI")) for f, freq in enumerate(frequency)] model_list = arlexecute.compute(model_list, sync=True) model_list = arlexecute.scatter(model_list) lprint("****** Starting dirty image calculation ******") start = time.time() dirty_list = invert_list_arlexecute_workflow(predicted_vis_list, model_list, vis_slices=vis_slices, context=context, facets=facets, use_serial_invert=use_serial_imaging, gcfcf=gcfcf_list) dirty, sumwt = arlexecute.compute(dirty_list, sync=True)[centre] end = time.time() results['time invert'] = end - start lprint("Dirty image invert took %.3f seconds" % (end - start)) lprint("Maximum in dirty image is %f, sumwt is %s" % (numpy.max(numpy.abs(dirty.data)), str(sumwt))) qa = qa_image(dirty) results['dirty_max'] = qa.data['max'] results['dirty_min'] = qa.data['min'] export_image_to_fits(dirty, "pipelines-arlexecute-timings-%s-dirty.fits" % context) # Corrupt the visibility for the GLEAM model lprint("****** Visibility corruption ******") start = time.time() corrupted_vis_list = corrupt_list_arlexecute_workflow(predicted_vis_list, phase_error=1.0, seed=seed) corrupted_vis_list = arlexecute.compute(corrupted_vis_list, sync=True) end = time.time() results['time corrupt'] = end - start lprint("Visibility corruption took %.3f seconds" % (end - start)) # Create the ICAL pipeline to run major cycles, starting selfcal at cycle 1. A global solution across all # frequencies (i.e. Visibilities) is performed. lprint("****** Starting ICAL ******") controls = create_calibration_controls() controls['T']['first_selfcal'] = 1 controls['T']['timescale'] = 'auto' start = time.time() ical_list = ical_list_arlexecute_workflow(corrupted_vis_list, model_imagelist=model_list, context=context, vis_slices=vis_slices, scales=[0, 3, 10], algorithm='mmclean', nmoment=3, niter=1000, fractional_threshold=0.1, threshold=0.01, nmajor=nmajor, gain=0.25, psf_support=64, deconvolve_facets=8, deconvolve_overlap=32, deconvolve_taper='tukey', timeslice='auto', global_solution=True, do_selfcal=True, calibration_context='T', controls=controls, use_serial_predict=use_serial_imaging, use_serial_invert=use_serial_imaging, gcfcf=gcfcf_list) end = time.time() results['time ICAL graph'] = end - start lprint("Construction of ICAL graph took %.3f seconds" % (end - start)) # Execute the graph start = time.time() result = arlexecute.compute(ical_list, sync=True) deconvolved, residual, restored, gaintables = result end = time.time() results['time ICAL'] = end - start lprint("ICAL graph execution took %.3f seconds" % (end - start)) qa = qa_image(deconvolved[centre]) results['deconvolved_max'] = qa.data['max'] results['deconvolved_min'] = qa.data['min'] deconvolved_cube = image_gather_channels(deconvolved) export_image_to_fits(deconvolved_cube, "pipelines-arlexecute-timings-%s-ical_deconvolved.fits" % context) qa = qa_image(residual[centre][0]) results['residual_max'] = qa.data['max'] results['residual_min'] = qa.data['min'] residual_cube = remove_sumwt(residual) residual_cube = image_gather_channels(residual_cube) export_image_to_fits(residual_cube, "pipelines-arlexecute-timings-%s-ical_residual.fits" % context) qa = qa_image(restored[centre]) results['restored_max'] = qa.data['max'] results['restored_min'] = qa.data['min'] restored_cube = image_gather_channels(restored) export_image_to_fits(restored_cube, "pipelines-arlexecute-timings-%s-ical_restored.fits" % context) # arlexecute.close() end_all = time.time() results['time overall'] = end_all - start_all lprint("At end, results are:") lprint(results) return results