拓展Homework 2的 R 语言练习，使用CHTC对一个未被发现的，引力透镜，高红移莱曼断裂星系进行全面搜索。
Requirement
This exercise, an expanded version of Homework 2, is a full-scale search using
the CHTC for an undiscovered, gravitationally lensed, high-redshift Lyman-
break galaxy.
Note: This homework requires running about 2500 jobs of a few minutes to an
hour each. This seems to take 2-3 hours when the CHTC is normally loaded. It
will become heavily loaded near this homework deadline, so that the
computation may take half a day or longer. Please plan accordingly. In
particular, it may not be realistic to try to finish this homework on the day
it is due.
Step 1
Revise your hw2.R from Homework 2 to a new hw4.R that takes two command-line
arguments: a template spectrum for which to search and a data directory in
which to find spectra to compare to the template. If called from the command
line without two arguments, it should display this message:
usage: Rscript hw4.R
An example usage is Rscript hw4.R cB58_Lyman_break.fit data.
Your hw4.R should write an output file whose name is the data directory name
followed by .csv and whose line format is distance,spectrumID,i where
- distance: your measure of the distance from this spectrum to the template
- spectrumID: the spectrum ID, e.g., spec-1353-53083-0579.fits
- i: the index in the spectrum at which your alignment with the template begins
The output file should include one line per spectrum in the data directory and
should be sorted by increasing distance. A sample line is
1032,spec-1353-53083-0579.fits,456, which says “the object in
spec-1353-53083-0579.fits has a distance 1032 from galaxy cB58 when red-
shifted by 456.” (If your hw2.R did not find spec-1353-53083-0579.fits among
its top three of the 100 spectra searched in HW2, revise your hw4.R further so
that it does find this spectrum from among those 100. Please ask early for
help with this if you need it.)
Step 2
Write a hw4.sh (modeled on myscript.sh that unpacks R, unpacks the FITSio
package, tells bash where R is, tells R where itself and its libraries are,
unpacks the current .tgz file (like 3586.tgz), and runs hw4.R on that
directory (like 3586).
Step 3
There are about 2.5 million spectra stored in the directory
/home/groups/STAT605/boss/tgz on the CHTC cluster. This directory contains
2459 .tgz files, each around 100 MB. Each .tgz file extracts to a directory
containing about 1000 spectra. The template cB58 is stored in
/home/groups/STAT605/boss/cB58_Lyman_break.fit.
- (a) Write an HTCondor submit script hw4_1job.sub that runs 1 job to processes the first .tgz file, 3586.tgz. That job should transfer 3586.tgz to a compute node, extract it to a directory 3586 containing 1000 .fits files, run hw4.R on the 3586 directory, and return the 3586.csv file described in (1) above. A few notes:
- Do not copy files from /home/groups/STAT605/boss/tgz to your /home/NetID directory, as this would unnecessarily blow up our disk space. Instead, refer to /home/groups/STAT605/boss/tgz in your .sub script and let HTCondor transfer the .tgz file to a compute node.
- I suggest that, while coding and debugging, you limit your hw4.R to process only about 3 of the 1000 spectra.
- If your computation is slow, so that processing 1000 spectra takes more than an hour, limit yourself to fewer than 1000 spectra per data directory. That is, ignore some of the spectra rather than run a very long job.
- (b) After your hw4_1job.sub runs correctly, note its Cpus, Disk (KB), and Memory (MB) use from the bottom of its .log file. Include these requirements (after increasing the Disk and Memory by a little) in your hw4_1job.sub script. Run it again, being sure to process all 1000 spectra.
Step 4
Write a hw4_5jobs.sub script that runs 5 parallel jobs to process the first
five .tgz files (3586.tgz, 3587.tgz, 3588.tgz, 3589.tgz, and 3590.tgz), one
per job.
- After your hw4_5jobs.sub runs correctly, note the Cpus, Disk (KB), and Memory (MB) use from the bottom of the five .log files. Include these requirements (after increasing the Disk and Memory by a little) in your hw4_5jobs.sub script. Run it again, being sure to process all 1000 spectra in each job.
- Write a hw4merge.sh script that merges your five .csv files into one sorted by distance and writes the best 100 spectra to hw4best100.csv.
Step 5
Write a submit script hw4.sub that runs 2459 parallel jobs to process all 2459
.tgz files, one per job. Run your hw4merge.sh again to merge your 2459 .csv
files into one, and write the best 100 spectra to hw4best100.csv.
Regarding this big run:
- Monitor your jobs with condor_q. To stop all your jobs, run condor_rm [NetID] (for me it’s condor_rm jgillett).
- No job should run longer than one hour. Kill any job that runs longer and redesign it.
Remember you may limit yourself to fewer than 1000 spectra per job, ignoring
some, to get job times under an hour. - Do not use more than 2 GB of data on any CHTC computer (i.e., no more than 2 GB per job).
- Do not launch a large number of jobs via an untested script. Start with 1 job, then 5, and only then 2459, as described above.
- Ask for help if you have trouble managing your jobs.
Step 6
Revise your hw2.Rmd to a new file hw4.Rmd (you may do this on the CHTC or on
your local machine):
- (a) Include your name and email address.
- (b) Include a leading summary paragraph describing what you did and mentioning any difficulties you encountered.
- (c) Your hw4.Rmd should read your hw4best100.csv file and make ten graphs, showing cB58 aligned with each of your top ten spectra from your search in hw4.sub. Include a legend with each graph identifying cB58 and the other spectrum. Reorder your graphs so that the best match (according to your eyes rather than your measure) is at the top. (d) Knit your hw4.Rmd to make hw4.html.
What to submit
Make a directory NetID/lyman. Copy only these files there:
- hw4.R
- hw4.sh
- hw4_1job.sub and a corresponding hw4_1job.log
- hw4_5jobs.sub and a corresponding hw4_5jobs.log (you can make this
via cat *.log > hw4_5jobs.login your log file directory) - hw4merge.sh
- hw4.sub and a corresponding hw4.log (you can make this via
cat *.log > hw4jobs.login your log file directory) - hw4best100.csv
- hw4.Rmd
- hw4.html
Include any supporting code files used by your scripts, but do not include the
data.
From the parent directory of NetID, run tar cvf NetID.tar NetID, Upload
NetID.tar to Canvas under the HW4 assignment. (I recommend downloading your
submitted file from Canvas into a temporary directory and confirming that it
is correct.)
Getting Help
Here are ways to get help.
