-
Notifications
You must be signed in to change notification settings - Fork 1
/
workflow.txt
28 lines (25 loc) · 1.34 KB
/
workflow.txt
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
workflow:
gather supervised rollouts:
open terminal, cd Izzy/vision-amt
run sudo xboxdrv --silent
open new tab,
run tpython TeleOp/ZekeTeleOp.py
When prompted, press r and place blocks at AR markers
press [esc] and use xbox controller to run. If center light is blinking, press [ctrl-c], run 5, restrat at 7
gather net test rollouts:
open terminal, cd Izzy/vision-amt
open saved_template_paths, and confirm that the paths are the test set (in ~/Izzy/vision-amt/data/amt/new_test_set.txt)
run tpython izzy_trainer.py
when prompted, press ['l']
set up and run
at end, click centers and press [esc] (possibly multiple times)
when prompted, press ['y'], ['n'] for success or failure
continue until no more data points
train a new net on supervised data:
open terminal, cd Izzy/vision-amt
run python scripts/merge_supervised.py [first training rollout number] [last training rollout number] -1
open Izzy/vision-amt/supervised_deltas.txt and move to Izzy/vision-amt/data/amt/deltas.txt
run python scripts/compile_supervised.py
open scripts/test_tf.py and confirm that net.optimize(300,data, batch_size=200) is uncommented (no path input)
run tpython scripts/test_tf.py
if more iterations are needed, enter in the path output at the end, and change the path variable, comment net.optimize(300,data, batch_size=200) and uncomment net.optimize(200,data,path = path, batch_size=200)