kengz/SLM-Lab

All Builds

update sac per spec to test #1998

kengz authored 5cb2cd90

complete00:01:21

Merge remote-tracking branch 'origin/master' into asyncsac #1997

kengz authored 03e31a70

complete00:00:35

Merge pull request #403 from kengz/radam #1996

Wah Loon Keng authored ddbc6230

complete00:00:23

revert spec changes #1995

kengz authored df5b79bd

complete00:00:23

add 6 worker spec #1994

kengz authored 7edbb9a4

complete00:00:22

use lr 3e-4 for all roboschool benchmark #1993

kengz authored ccf91e2e

complete00:03:28

Merge pull request #402 from kengz/robobench #1992

Wah Loon Keng authored 4f7287d2

complete00:01:50

fix import #1991

kengz authored 6d3b7a29

complete00:00:37

Merge pull request #402 from kengz/robobench #1990

Wah Loon Keng authored 4f7287d2

complete00:00:22

remove old classic benchmark table #1989

kengz authored aa9222c0

complete00:00:17

update roboschool benchmark table #1988

kengz authored da1f3a2a

complete00:00:21

kengz authored 8318059a

complete00:00:22

lower lr for sac humanoid #1986

kengz authored 35fad887

complete00:00:40

use wider net for ppo #1985

kengz authored 51590120

complete00:00:23

kengz authored 7b0cfe9c

complete00:00:21

lower clip grad for ac fam #1983

kengz authored 5c3bc8e7

complete00:01:05

tune ac fam #1982

kengz authored 469b1233

complete00:01:15

tune ac with higher lr #1981

kengz authored b2a5d8ae

complete00:00:30

run sac roboschool for 2m frames #1980

kengz authored b7f6de84

complete00:00:24

lower ac lr, increase training freq #1979

kengz authored 839922b4

complete00:00:23

split humanoid specs #1978

kengz authored 872b50e0

complete00:00:23

split humanoid specs into easy hard #1977

kengz authored d0d94d44

complete00:00:21

try 16 envs for ac fam #1976

kengz authored 412075b8

complete00:03:24

try 4 envs for ac fam #1975

kengz authored 836eeaac

complete00:00:26

tune ac lr #1974

kengz authored dfa82e84

complete00:00:22

lower clip grad val to 1.0, restore ac freq and arch #1973

kengz authored 7081f7b4

complete00:00:33

wider arch for ac fam #1972

kengz authored 568cf1b7

complete00:00:26

Merge pull request #401 from kengz/fix-analysis #1971

Wah Loon Keng authored 90d885d7

complete00:00:23

Merge pull request #399 from kengz/sac2 #1970

Wah Loon Keng authored 4fb2efed

complete00:00:21

fix retro analysis experiment scan criteria #1969

kengz authored cd673b31

complete00:00:21

Merge pull request #399 from kengz/sac2 #1968

Wah Loon Keng authored 4fb2efed

complete00:00:21

bump version #1967

kengz authored f6a2922b

complete00:00:17

correct benchmark image ordering #1966

kengz authored 866ab7b5

complete00:00:20

update SAC benchmark table #1965

kengz authored 7c935da6

complete00:00:17

expand action fc layers #1964

kengz authored 0b130a07

complete00:00:19

optimize GumbelSoftmax output shape to speedup algo #1963

kengz authored 6ddae89e

complete00:00:21

put log_alpha on the same device #1962

kengz authored b210320f

complete00:00:17

revert temp back to 1.0 #1961

kengz authored 320721a3

complete00:00:27

change sac_per_hopper to halfcheetah #1960

kengz authored f5eea491

complete00:00:29

default gumbelsoftmax temperature 0.1 #1959

kengz authored 5501dbfc

complete00:00:18

clarify action sampling comment #1958

kengz authored 9ee0f06e

complete00:00:23

restore multivariate #1957

kengz authored 615ef42b

complete00:00:21

use a single log prob calc for cont correctly #1956

kengz authored 64e3ce49

complete00:00:18

make net API generic by making scheduler and clock optional #1955

kengz authored 1ba572ed

complete00:00:41

restore F import #1954

kengz authored 4b954f03

complete00:00:53

fix gumbel unit test #1953

kengz authored 49e9fbbd

complete00:00:17

sum external log prob too #1952

kengz authored 5b1ec1eb

complete00:00:25

testrun multiple Normal distributions for SAC #1951

kengz authored c32887e2

complete00:00:52

remove extra import #1950

kengz authored b8696fae

complete00:00:17

update sac tests #1949

kengz authored 2a9b330b

complete00:02:47