Closed belledon closed 3 years ago
For each batch create a condlist.json such that there are 6 conditions (tracker x probe). We are aiming for 120 trials per condition. For each scene, there will be 10 repetitions (1 tracker x 5 time points x 2 target/distractor) per condition
Does this mean that, say, Alice will see 12 scenes with tracker 1 and all with probe, and Bob will see 12 scenes with tracker 2, but no probe whatsoever?
n o sorry, this means that alice sees scene 1 tracker 1 w/ probe scene 2 tracker 2 no probe and bob sees scene 1 tracker 2 w/ probe scene 2 tracker 2 no probe....
imagine there where only 2 scenes, 2 trackers, 2 time points the break down would be
condition 1
scene | probe | tracker | time | query |
---|---|---|---|---|
1 | T | 1 | 1 | Trg |
1 | T | 1 | 1 | Dis |
1 | T | 1 | 2 | Trg |
1 | T | 1 | 2 | Dis |
2 | F | 2 | 1 | Trg |
2 | F | 2 | 1 | Dis |
2 | F | 2 | 2 | Trg |
2 | F | 2 | 2 | Dis |
condition 2
scene | probe | tracker | time | query |
---|---|---|---|---|
1 | T | 2 | 1 | Trg |
1 | T | 2 | 1 | Dis |
1 | T | 2 | 2 | Trg |
1 | T | 2 | 2 | Dis |
2 | F | 1 | 1 | Trg |
2 | F | 1 | 1 | Dis |
2 | F | 1 | 2 | Trg |
2 | F | 1 | 2 | Dis |
condition 3
scene | probe | tracker | time | query |
---|---|---|---|---|
1 | F | 1 | 1 | Trg |
1 | F | 1 | 1 | Dis |
1 | F | 1 | 2 | Trg |
1 | F | 1 | 2 | Dis |
2 | T | 2 | 1 | Trg |
2 | T | 2 | 1 | Dis |
2 | T | 2 | 2 | Trg |
2 | T | 2 | 2 | Dis |
condition 2
scene | probe | tracker | time | query |
---|---|---|---|---|
1 | F | 2 | 1 | Trg |
1 | F | 2 | 1 | Dis |
1 | F | 2 | 2 | Trg |
1 | F | 2 | 2 | Dis |
2 | T | 1 | 1 | Trg |
2 | T | 1 | 1 | Dis |
2 | T | 1 | 2 | Trg |
2 | T | 1 | 2 | Dis |
as of now there are a total of 24 scenes with 12 conditions: batch (2) x probe (2) x tracker (3) However, we will be splitting the 24 scenes along batch such that we can run 2 experiments of 12 trials with 6 (tracker x probe)
You can find an early draft of render under
probe_stimuli/probes.tar.gz
and the probe map underprobe_stimuli/attention_analysis/exp0_probe_map.csv
.The file name breaks down to
<scene>_<tracker>_t_<epoch>_[probe|noprobe]_[trg|dis].mp4
Tasks
[x] Split the 24 scenes such that each batch of 12 is balanced across scene number (4 -> 118, something like every other).
[x] For each batch create a condlist.json such that there are 6 conditions (tracker x probe). We are aiming for 120 trials per condition. For each scene, there will be 10 repetitions (1 tracker x 5 time points x 2 target/distractor) per condition
[ ] find a way to rotate each repetition to prevent memorization. One way is to change the render script but I would like to explore somehting on the psiturk side to see if its easy (otherwise we can change rendering) see link. Note that the same rotations can be used across trackers and scenes (divide 2pi by 10) since each subject (condition) only sees one scene-tracker pair per scene.
[ ] add those box paths above to setup datasets