r - Finding Overlaps between interval sets / Efficient Overlap Joins

Question

Overview:

I need to join two tables:

ref contains the time intervals (from t1 to t2) along with an id for each interval and a space where this interval occurs.

map contains time intervals (t1 to t2) each with a result res and its corresponding space.

I wish to obtain/join all intervals (and their score) of map that fall within the intervals in ref onto ref.

example:

ref <- data.table(space=rep('nI',3),t1=c(100,300,500),t2=c(150,400,600),id=letters[1:3])

map <- data.table(space=rep('nI',241),t1=seq(0,1200,by=5),t2=seq(5,1205,by=5),res=rnorm(241))

they look like:

> ref
space  t1  t2 id
1:    nI 100 150  a
2:    nI 300 400  b
3:    nI 500 600  c

> map
space   t1   t2        res
1:    nI    0    5 -0.7082922
2:    nI    5   10  1.8251041
3:    nI   10   15  0.2076552
4:    nI   15   20  0.8047347
5:    nI   20   25  2.3388920
---                           
237:    nI 1180 1185  1.0229284
238:    nI 1185 1190 -0.3657815
239:    nI 1190 1195  0.3013489
240:    nI 1195 1200  1.2947271
241:    nI 1200 1205 -1.5050221

(UPDATE) Solution

• ?data.table::foverlaps is the key here.

I need to join all the map intervals that occur "within" the intervals of ref and I am not interested in intervals that do not match this key so use nomatch=0L.

setkey(ref,space,t1,t2)

foverlaps(map,ref,type="within",nomatch=0L)

which gives:

space  t1  t2 id i.t1 i.t2         res
1:    nI 100 150  a  100  105 -0.85202726
2:    nI 100 150  a  105  110  0.79748876
3:    nI 100 150  a  110  115  1.49894097
4:    nI 100 150  a  115  120  0.47719957
5:    nI 100 150  a  120  125 -0.95767896
6:    nI 100 150  a  125  130 -0.51054673
7:    nI 100 150  a  130  135 -0.08478700
8:    nI 100 150  a  135  140 -0.69526566
9:    nI 100 150  a  140  145  2.14917623
10:    nI 100 150  a  145  150 -0.05348163
11:    nI 300 400  b  300  305  0.28834548
12:    nI 300 400  b  305  310  0.32449616
13:    nI 300 400  b  310  315  1.16107248
14:    nI 300 400  b  315  320  1.08550676
15:    nI 300 400  b  320  325  0.84640788
16:    nI 300 400  b  325  330 -2.15485447
17:    nI 300 400  b  330  335  1.59115714
18:    nI 300 400  b  335  340 -0.57588128
19:    nI 300 400  b  340  345  0.23957563
20:    nI 300 400  b  345  350 -0.60824259
21:    nI 300 400  b  350  355 -0.84828189
22:    nI 300 400  b  355  360 -0.43528701
23:    nI 300 400  b  360  365 -0.80026281
24:    nI 300 400  b  365  370 -0.62914234
25:    nI 300 400  b  370  375 -0.83485164
26:    nI 300 400  b  375  380  1.46922713
27:    nI 300 400  b  380  385 -0.53965310
28:    nI 300 400  b  385  390  0.98728765
29:    nI 300 400  b  390  395 -0.66328893
30:    nI 300 400  b  395  400 -0.08182384
31:    nI 500 600  c  500  505  0.72566100
32:    nI 500 600  c  505  510  2.27878366
33:    nI 500 600  c  510  515  0.72974139
34:    nI 500 600  c  515  520 -0.35358019
35:    nI 500 600  c  520  525 -1.20697646
36:    nI 500 600  c  525  530 -0.01719057
37:    nI 500 600  c  530  535  0.06686472
38:    nI 500 600  c  535  540 -0.40866088
39:    nI 500 600  c  540  545 -1.02697573
40:    nI 500 600  c  545  550  2.19822065
41:    nI 500 600  c  550  555  0.57075648
42:    nI 500 600  c  555  560 -0.52009726
43:    nI 500 600  c  560  565 -1.82999177
44:    nI 500 600  c  565  570  2.53776578
45:    nI 500 600  c  570  575  0.85626293
46:    nI 500 600  c  575  580 -0.34245708
47:    nI 500 600  c  580  585  1.21679869
48:    nI 500 600  c  585  590  1.87587020
49:    nI 500 600  c  590  595 -0.23325264
50:    nI 500 600  c  595  600  0.18845022
space  t1  t2 id i.t1 i.t2         res

Answer 1

Ha, nice timing :). Just a few days back, overlap joins (or interval joins) was implemented. in data.table The function is foverlaps() and is available from the github project page. Make sure to have a look at ?foverlaps.

setkey(ref, space, t1, t2)
foverlaps(map, ref, type="within", nomatch=0L)

I think this is what you're after. This'll result in the join result only where there's a match, and it'll check for t1,t2 overlaps between ref and map within space identifier.. If you don't want that, just remove space from the key column. And if you want all matches, remove nomatch=0L - the default is nomatch=NA which returns all.

The function is new (but has been rigorously tested) and is therefore not feature complete. If you've any suggestions for improvement or come across any issues, please feel free to file an issue.