Parameters

The following tables show the optimal parameters obtained by grid search on validation dataset.

NHQ-NPG_kgraph

Please see here for more detailed meaning about these parameters.

	SIFT1M	GIST1M	GloVe	Crawl	Audio	Msong	Enron	UQ-V	Paper
K	100	100	100	100	100	400	100	100	100
L	100	100	100	100	130	420	100	100	100
iter	12	12	12	12	12	12	12	12	12
S	10	10	10	10	10	10	10	15	10
R	300	300	100	100	100	300	100	200	100
RANGE	20	20	20	20	20	80	20	20	20
PL	350	50	150	150	250	150	50	150	50
B	0.4	0.6	0.2	0.4	0.6	0.6	0.2	0.6	0.2
M	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0

Please see here for more detailed meaning about these parameters.

	SIFT1M	GIST1M	GloVe	Crawl	Audio	Msong	Enron	UQ-V	Paper
MaxM0	40	20	20	100	60	50	90	30	90
efConstruction	100	200	900	500	800	900	900	300	400

Please see here for more detailed meaning about these parameters.

	SIFT1M	GIST1M	GloVe	Crawl
K	200	100	400	400
L	220	130	420	400
iter	12	12	12	12
S	25	20	25	20
R	200	300	300	200
RANGE	40	40	90	90
PL	50	350	50	350
B	1.0	1.0	0.8	1.0
M	1.0	1.0	1.0	1.0

Please see here for more detailed meaning about these parameters.

dataset	SIFT1M	GIST1M	GloVe	Crawl
MaxM0	40	60	60	60
efConstruction	300	300	700	300

When performing a hybrid query, we set $\omega _{x}=1$ and $\omega _{y}=$weight_search as follows.

	SIFT1M	GIST1M	GloVe	Crawl	Audio	Msong	Enron	UQ-V	Paper
weight_search	140000	9	80	90	3e+10	5500	480000	2	5000