Differences
This shows you the differences between two versions of the page.
|
talksposters:spsearch [2024/04/29 10:04] lspitler created |
talksposters:spsearch [2024/04/29 11:18] (current) lspitler |
||
|---|---|---|---|
| Line 5: | Line 5: | ||
| The tutorial uses ''PRESTO''. Although ''PRESTO'' may not be the best for a computationally intensive pipelines, it is pedagogically useful by allowing the user to manually run each pipeline step. | The tutorial uses ''PRESTO''. Although ''PRESTO'' may not be the best for a computationally intensive pipelines, it is pedagogically useful by allowing the user to manually run each pipeline step. | ||
| + | [[https://github.com/scottransom/presto/blob/master/docs/PRESTO_search_tutorial.pdf | Link to PRESTO tutorial in PRESTO's github repo]] | ||
| + | |||
| + | Notes: | ||
| + | * I providing the commands outside of singularity. Run them with singularity as you prefer. | ||
| + | * I assume that the input data is in filterbank format and done my best to note where commands may differ in the case of PSRFITS format. | ||
| + | |||
| + | ===== What you need ===== | ||
| + | |||
| + | - Data: short observation of a test pulsar (MY_DATA.fil) | ||
| + | - Software: either a working ''PRESTO'' install or a singularity image containing ''PRESTO'' | ||
| + | |||
| + | ===== Simple SP pipeline ===== | ||
| + | |||
| + | === Step 1: inspect data === | ||
| + | |||
| + | Become familiar with the structure and contents of the data using ''waterfaller.py''. | ||
| + | |||
| + | Example command: | ||
| + | |||
| + | ''> waterfaller.py %%--%%show-ts %%--%%show-spec %%--%%colour-map='viridis' -d 57 -T 100 -t 5 %%--%%downsamp=8 %%--%%scaleindep MY_DATA.fil'' | ||
| + | |||
| + | Parameters to play around with: | ||
| + | * Turn on and off dedispersion | ||
| + | * Go to a different time in the file | ||
| + | * Change the downsample factor | ||
| + | * Look at data with and without independently scaling each frequency channel | ||
| + | |||
| + | === Step 2: apply RFI mitigation === | ||
| + | |||
| + | ''PRESTOs'' removes RFI by calculating a mask. The mask informtion is saved in a number of files, one of which can be passed to the following steps. Therefore, a new, cleaned copy of the original data file is not produced. | ||
| + | |||
| + | Example command: | ||
| + | |||
| + | ''> rfifind -o mask_file_name -time 1.0 -zapchan N:N MY_DATA.fil'' | ||
| + | |||
| + | Parameters to play around with: | ||
| + | * Duration of processing time (-time) | ||
| + | * In the case of PSRFITS data: -noscales -nooffsetts | ||
| + | * Change thresholding parameters (-timesig, -freqsig, -chanfrac, -intfrac) | ||
| + | * Add channels to manually zap (-zapchan) | ||
| + | |||
| + | Note, ''rfifind'' can be quite slow to run, but it is possible under some circumstances to recalculate the mask from the mask itself rather than rerunning the entire code. | ||
| + | |||
| + | Data file: MY_DATA.fil \\ | ||
| + | Original mask: mask_file_name_rfifind.mask | ||
| + | |||
| + | ''> rfifind -o mask_file_name -nocompute -freqsig 8 -mask mask_file_name_rfifind.mask MY_DATA.fil'' | ||
| + | |||
| + | === Step 3: generate dedispersed time series === | ||
| + | |||
| + | This command assumes the data are for B0355+54 - change the DM range accordingly if using a different test pulsar. | ||
| + | |||
| + | This command can be used to generate a single dedipsersed time series by setting -numdms 1. Having dedispered time series over a range around the true value is useful for in Step 4. | ||
| + | |||
| + | ''> prepsubband -nobary -lodm 52.4 -dmstep 0.5 -numdms 20 -clip 0 -downsamp 1 -mask mask_file_name_rfifind.mask -o my_ts MY_DATA.fil'' | ||
| + | |||
| + | Variations: | ||
| + | * PSRFITS files: if you are searching a PSRFITS file and applied ''%%--%%noscales %%-nooffsets%%'' in the ''RFIfind'' stage, you should do so here too. | ||
| + | * Change the DM step and DM range (and see its impact on Step 4) | ||
| + | * Change the downsampling factor | ||
| + | * Run with clipping to see if this changes the time series | ||
| + | |||
| + | Manually inspect the time series: | ||
| + | |||
| + | ''> exploredat my_ts.dat'' | ||
| + | |||
| + | * Can you see bright pulses? | ||
| + | * Note the slowly varying | ||
| + | * If you've applied an RFIfind mask, what impact does this have on the time series? | ||
| + | |||
| + | === Step 4: generate single pulse candidates === | ||
| + | |||
| + | ''> single_pulse_search.py -m 30 -t 6 -b *.dat'' | ||
| + | |||
| + | Parameters to vary: | ||
| + | * Turn on bad block flagging | ||
| + | * Use different detrend chunksizes | ||
| + | |||
| + | Look at SP diagnostic plot: my_ts_singlepulse.ps | ||
| + | |||
| + | Recreate this plot with a different threshold or time range: | ||
| + | |||
| + | ''> single_pulse_search.py -m 30 -t 8 -s 0 -e 120 -b *.singlepulse'' | ||
| + | |||
| + | === Step 5: machine learning classifier === | ||
| + | |||
| + | To be done... | ||
