factmsieve filtering too often?

Belteshazzar · 2020-07-02, 19:35

Am doing a C138 using factmsieve.py. Once it hit 100% of the estimated minimum relations, it's running filtering after every single batch of sieving, even though the number of relations it gets from one batch (~160000, ~0.8% of the estimated minimum) is likely nowhere near enough to be done sieving.

Since the batches take just over 10 minutes each and the filtering takes nearly 6 minutes, this seems to be making things take more than one and a half times as long as necessary. Why is it doing this?

(after ten rounds of this ergo almost an hour of wasted filtering effort it's now for the first time showing relations > ideals, getting closer...)

Belteshazzar · 2020-07-02, 22:28

It was able to build a matrix on the 13th filtering try, got the p60+p79; the wasted filtering efforts spent about an hour and a half of the ~24h spent on the factorization. Then it tripped on the finish line:

Code:

-> Computing time scale for this machine...
-> procrels -speedtest> PIPE
Traceback (most recent call last):
  File "$DIR\GGNFS\factmsieve-0.86.py", line 2253, in <module>
    output_summary(NAME, fact_p, pols_p, poly_p, lats_p)
  File "$DIR\GGNFS\factmsieve-0.86.py", line 1911, in output_summary
    (ret, res) = run_exe(PROCRELS, '-speedtest', out_file = subprocess.PIPE)
  File "$DIR\GGNFS\factmsieve-0.86.py", line 350, in run_exe
    p = subprocess.Popen([ex] + args.split(' '), **al)
  File "$PYDIR\Python37\lib\subprocess.py", line 775, in __init__
    restore_signals, start_new_session)
  File "$PYDIR\Python37\lib\subprocess.py", line 1178, in _execute_child
    startupinfo)
FileNotFoundError: [WinError 2] The system cannot find the file specified

(where I've replaced some directory names with obv.vars)

Batalov · 2020-07-03, 00:00

Get the procrels binary, then. It is completely unnecessary though.
Option 2: Comment that line out.

VBCurtis · 2020-07-03, 00:25

There are a couple of things you can do to modify factmsieve to reduce the time wasted.
The most obvious is to edit the relations-wanted count to more accurately represent the number needed- such that it filters just once, perhaps twice.

The less obvious but possibly also useful is to set mfb to be 2* lpb - 4 in the parameters-setting code (rather than 2 * lpb -1); that will reduce the relations needed quite a bit, and may also speed the overall factorization. This also has the benefit of helping for all LP choices and all sizes of factorization, rather than having to run a job at each LP size before adjusting the formula (or adjusting it poorly- I spent dozens of jobs adjusting those rels-wanted formulae back before I moved to CADO).

2020-07-02, 19:35	#1
Belteshazzar Feb 2011 2⁵ Posts	factmsieve filtering too often? Am doing a C138 using factmsieve.py. Once it hit 100% of the estimated minimum relations, it's running filtering after every single batch of sieving, even though the number of relations it gets from one batch (~160000, ~0.8% of the estimated minimum) is likely nowhere near enough to be done sieving. Since the batches take just over 10 minutes each and the filtering takes nearly 6 minutes, this seems to be making things take more than one and a half times as long as necessary. Why is it doing this? (after ten rounds of this ergo almost an hour of wasted filtering effort it's now for the first time showing relations > ideals, getting closer...)

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
factmsieve.py Questions	EdH	Msieve	3	2016-11-22 17:47
I am new Running Factmsieve i am Stuck please help	zukhruf	Msieve	2	2015-11-25 12:58
Finding factmsieve.py	ThomRuley	Msieve	2	2015-06-27 04:23
I am very new to factoring and factmsieve	Hailstone	Msieve	10	2015-01-08 03:34
Factmsieve.py acting strange	ThomRuley	Msieve	11	2013-11-26 02:38

2020-07-03, 00:00	#3
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 9,901 Posts	Get the procrels binary, then. It is completely unnecessary though. Option 2: Comment that line out.

2020-07-03, 00:25	#4
VBCurtis "Curtis" Feb 2005 Riverside, CA 2²×17×79 Posts	There are a couple of things you can do to modify factmsieve to reduce the time wasted. The most obvious is to edit the relations-wanted count to more accurately represent the number needed- such that it filters just once, perhaps twice. The less obvious but possibly also useful is to set mfb to be 2* lpb - 4 in the parameters-setting code (rather than 2 * lpb -1); that will reduce the relations needed quite a bit, and may also speed the overall factorization. This also has the benefit of helping for all LP choices and all sizes of factorization, rather than having to run a job at each LP size before adjusting the formula (or adjusting it poorly- I spent dozens of jobs adjusting those rels-wanted formulae back before I moved to CADO).