biochem-fan
commented
8 years ago Pr-SAD dataset, 10K lattices cppxfel 1b7e98c
-
test1: default 25.51, 27.92, 29.15, 29.61, 29.57, 29.62
-
test2: INITIAL_RLP_SIZE 0.0002, STEP_SIZE_RLP_SIZE 0.00004 27.65, 29.58, 31.63, 32.92, 33.56, 33.34
-
test3: MINIMIZATION_METHOD 0 23.86, 26.69, 27.18, 27.09, 26.58, 26.16
-
test4: SCALING_STRATEGY 4 26.25, 27.53, 28.99, 28.28, 28.19, 27.58
-
test5: INITIAL_RLP_SIZE 0.0003, STEP_SIZE_RLP_SIZE 0.00004 27.75, 27.02, 29.78, 31.95, 33.03, 33.45
-
test6: test2 + STEP_SIZE_ORIENTATION 0.02 29.03, 31.38, 33.36, 34.24, 34.22, 34.26
-
test7: test5 + STEP_SIZE_ORIENTATION 0.02 29.49, 28.48, 31.65, 32.65, 33.28, 33.73
-
test8: test6 + SCALING_STRATEGY 4 29.46, 31.01, 32.25, 32.35, 33.00, 32.34
-
test9: test6 + CORRELATION_THRESHOLD 0.6 28.92, 32.01, 33.06, 33.80, 33.72, 33.85
-
test10: test6 + TARGET_FUNCTION 11 28.00, 30.18, 32.74, 33.56, 33.77, 33.90
-
process_hkl: 24.91
-
process_hkl --scale: 30.80
-
partialator: 32.75
-
cxi.merge rs2, ref=CrystFEL: 35.89, 35.59
-
cxi.merge rs2, ref=F_model: 34.19, 35.54, 34.19
-
PRIME: 28.14, , , *, ,18.13
hmeduyvesteyn
It seems that there are at least two problems. Sigma values are for Hg-SAD datasets (11,000 images).
Regression between Apr 26 and May 7
04701bf (Apr 26) is OK (15.66 then 25.17 sigma in cycle 0 and 1). b7143d4 (Apr 28) is OK (15.70 then 25.30) 2b2f4f5 (May 7) is not (13.46 then 21.31 sigma).
Bug in initial merging, introduced between May 7 and Jul 16
At commit 27027bb (Jul 16), the completeness was very low (~ 40%) because most images were rejected as having low CC. ANODE peak heights were < 10 sigma. In cycle 1, 90% of images were rejected. Usually, I get < 40 % rejected at cycle 1 and rejected images are recovered in later cycles as the internal reference improves.
With MC integrated intensities from CrystFEL as the INITIAL_MTZ, I got 22.26 sigma at cycle 0 with completeless 98%. Sigma values hardly improved in later cycles.
27027bb using reference from b7143d4 (Apr 28): 18.00 then 21.53 sigma