forked from hpclab/quickrank
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathCHANGES
34 lines (27 loc) · 1.21 KB
/
CHANGES
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
2015 Sep. 8.
- Cleaning up of part of the code related to document scoring
2015 Mar. 12.
- Implementation of CoordinateAscent
Metzler, D., Croft, W.B.: Linear feature-based models for information retrieval.
Information Retrieval 10(3), 257–274 (2007)
2015 Feb. 10
- Removed histagrams of labels squares
Standard Experiment:
time ./bin/quicklearn --train tests/data/msn1.fold1.train.5k.txt --valid tests/data/msn1.fold1.vali.5k.txt --test tests/data/msn1.fold1.test.5k.txt --num-trees 500 --end-after-rounds 0
# NEW
Training Time: 26.66 s.
NDCG@10 on test data = 0.3353
# OLD
Training Time: 38.70 s.
NDCG@10 on test data = 0.3353
2015 Jan. 8
- Instrumented code: counting number of traversed node during test.
2014 Dec. 18
- Trying to review OpenMP use.
Standard Experiment:
time ./bin/quicklearn --train tests/data/msn1.fold1.train.5k.txt --valid tests/data/msn1.fold1.vali.5k.txt --test tests/data/msn1.fold1.test.5k.txt --num-trees 100
NDCG@10 on test data = 0.3449
10.81s user 0.86s system 121% cpu 9.621 total
2014 Dec. 18
- Quick-sort and Merge-Sort Implementation are always slower than std::sort.
These have been removed. Only radix sort is maintained for the initial large input.