final atari running settings

Author: d9w

cfg/base.yaml

@@ -1,24 +1,24 @@
 # General settings
 starting_nodes: 50
-static_node_size: 100
+static_node_size: 40
 node_size_cap: 150
 node_inputs: 2
-save_best: False
+save_best: True
 weights: True
 mutate_method: ":gene_mutate"
 crossover_method: ":single_point_crossover"
-recurrency: 0.0
-total_evals: 2000
+recurrency: 0.1
+total_evals: 10000
 input_start: -1.0
 # Mutation
-active_mutate: True
+active_mutate: False
 input_mutation_rate: 0.1
-output_mutation_rate: 0.1
+output_mutation_rate: 0.6
 node_mutation_rate: 0.1
 node_size_delta: 0.1
 modify_mutation_rate: 0.25
 # Oneplus settings
-lambda: 5
+lambda: 9
 # GA settings
 ga_population: 50
 ga_elitism_rate: 0.1

experiments/atari.jl

@@ -7,14 +7,27 @@ function play_atari(c::Chromosome, game::Game, id::String)
     reset_game(game.ale)
     reward = 0.0
     frames = 0
+    p_action = game.actions[1]
+    act_count = 0
     while ~game_over(game.ale)
         output = process(c, get_rgb(game))
         action = game.actions[indmax(output)]
-        for i in 1:4
+#       if action == p_action
+#           act_count += 1
+#       else
+#           p_action = action
+#           act_count = 0
+#       end
+#       if act_count > 500
+#           println("Termination due to repetitive action ", id)
+#           return -Inf
+#       end
+        reward += act(game.ale, action)
+        frames += 1
+        if rand() < 0.25
             reward += act(game.ale, action)
-            frames += 1
         end
-        if frames > 5000
+        if frames > 18000
             println("Termination due to frame count on ", id)
             break
         end
@@ -37,10 +50,10 @@ function get_args()
         default = "qbert"
         "--ea"
         arg_type = String
-        required = true
+        default = "oneplus"
         "--chromosome"
         arg_type = String
-        required = true
+        default = "CGPChromo"
     end
 
     CGP.Config.add_arg_settings!(s)
@@ -64,6 +77,6 @@ nin = 3 # r g b
 nout = length(game.actions)
 fit = x->play_atari(x, game, args["id"])
 
-maxfit, best = ea(ctype, nin, nout, fit; seed=args["seed"])
+maxfit, best = ea(ctype, nin, nout, fit; seed=args["seed"], id=args["id"])
 close!(game)
 Logging.info(@sprintf("E%0.6f", -maxfit))

scripts/.Rhistory

@@ -0,0 +1,11 @@
+install.packages("Rmpi")
+q()
+install.packages('Rmpi', config.args='--with-R-mpi-include=/usr/local/intel/impi/4.1.3.049/intel64/include --with-Rmpi-libpath=/usr/local/intel/impi/4.1.3.049/intel64/lib')
+install.packages('Rmpi', config.args='--with-R-mpi-include=/usr/local/intel/impi/4.1.3.049/intel64/include --with-Rmpi-libpath=/usr/local/intel/impi/4.1.3.049/intel64/lib')
+help(.libPaths())
+.libPaths()
+.libPaths("/usr/local/intel/impi/4.1.3.049/intel64/lib")
+.libPaths()
+install.packages('Rmpi')
+install.packages('Rmpi', config.args=c("--with-R-mpi-include=/usr/local/intel/impi/4.1.3.049/intel64/include","--with-Rmpi-libpath=/usr/local/intel/impi/4.1.3.049/intel64/lib"))
+q()

scripts/atari.cfg

@@ -0,0 +1,4 @@
+0-9 julia atari.jl --seed %o --log boxing_%o.log --id boxing
+10-19 julia atari.jl --seed %o --log centipede_%o.log --id centipede
+20-29 julia atari.jl --seed %o --log air_raid_%o.log --id air_raid
+30-39 julia atari.jl --seed %o --log pong_%o.log --id pong

scripts/batch.sh

@@ -1,10 +1,18 @@
 #!/bin/sh
 #SBATCH -J PCGP
-#SBATCH -N 1
-#SBATCH -n 20
+#SBATCH -N 32
+#SBATCH -n 640
 #SBATCH --ntasks-per-node=20
 #SBATCH --ntasks-per-core=1
 #SBATCH [email protected]
 #SBATCH --mail-type=ALL
 
-srun ./run.sh
+export CGP=/users/p16043/wilson/CGP.jl
+export WORK_DIR=/tmpdir/wilson/dennis/$SLURM_JOB_ID
+
+mkdir -p $WORK_DIR
+cp -r $CGP/cfg $WORK_DIR/
+cp -r $CGP/experiments/atari.jl $WORK_DIR/
+cd $WORK_DIR
+
+srun --multi-prog $CGP/scripts/ids.cfg

scripts/ids.cfg

@@ -0,0 +1,64 @@
+0-9 julia atari.jl --seed %o --log air_raid_%o.log --id air_raid
+10-19 julia atari.jl --seed %o --log alien_%o.log --id alien
+20-29 julia atari.jl --seed %o --log amidar_%o.log --id amidar
+30-39 julia atari.jl --seed %o --log assault_%o.log --id assault
+40-49 julia atari.jl --seed %o --log asterix_%o.log --id asterix
+50-59 julia atari.jl --seed %o --log asteroids_%o.log --id asteroids
+60-69 julia atari.jl --seed %o --log atlantis_%o.log --id atlantis
+70-79 julia atari.jl --seed %o --log bank_heist_%o.log --id bank_heist
+80-89 julia atari.jl --seed %o --log battle_zone_%o.log --id battle_zone
+90-99 julia atari.jl --seed %o --log beam_rider_%o.log --id beam_rider
+100-109 julia atari.jl --seed %o --log berzerk_%o.log --id berzerk
+110-119 julia atari.jl --seed %o --log bowling_%o.log --id bowling
+120-129 julia atari.jl --seed %o --log boxing_%o.log --id boxing
+130-139 julia atari.jl --seed %o --log breakout_%o.log --id breakout
+140-149 julia atari.jl --seed %o --log carnival_%o.log --id carnival
+150-159 julia atari.jl --seed %o --log centipede_%o.log --id centipede
+160-169 julia atari.jl --seed %o --log chopper_command_%o.log --id chopper_command
+170-179 julia atari.jl --seed %o --log crazy_climber_%o.log --id crazy_climber
+180-189 julia atari.jl --seed %o --log defender_%o.log --id defender
+190-199 julia atari.jl --seed %o --log demon_attack_%o.log --id demon_attack
+200-209 julia atari.jl --seed %o --log double_dunk_%o.log --id double_dunk
+210-219 julia atari.jl --seed %o --log elevator_action_%o.log --id elevator_action
+220-229 julia atari.jl --seed %o --log enduro_%o.log --id enduro
+230-239 julia atari.jl --seed %o --log fishing_derby_%o.log --id fishing_derby
+240-249 julia atari.jl --seed %o --log freeway_%o.log --id freeway
+250-259 julia atari.jl --seed %o --log frostbite_%o.log --id frostbite
+260-269 julia atari.jl --seed %o --log gopher_%o.log --id gopher
+270-279 julia atari.jl --seed %o --log gravitar_%o.log --id gravitar
+280-289 julia atari.jl --seed %o --log hero_%o.log --id hero
+290-299 julia atari.jl --seed %o --log ice_hockey_%o.log --id ice_hockey
+300-309 julia atari.jl --seed %o --log jamesbond_%o.log --id jamesbond
+310-319 julia atari.jl --seed %o --log journey_escape_%o.log --id journey_escape
+320-329 julia atari.jl --seed %o --log kaboom_%o.log --id kaboom
+330-339 julia atari.jl --seed %o --log kangaroo_%o.log --id kangaroo
+340-349 julia atari.jl --seed %o --log krull_%o.log --id krull
+350-359 julia atari.jl --seed %o --log kung_fu_master_%o.log --id kung_fu_master
+360-369 julia atari.jl --seed %o --log montezuma_revenge_%o.log --id montezuma_revenge
+370-379 julia atari.jl --seed %o --log ms_pacman_%o.log --id ms_pacman
+380-389 julia atari.jl --seed %o --log name_this_game_%o.log --id name_this_game
+390-399 julia atari.jl --seed %o --log phoenix_%o.log --id phoenix
+400-409 julia atari.jl --seed %o --log pitfall_%o.log --id pitfall
+410-419 julia atari.jl --seed %o --log pong_%o.log --id pong
+420-429 julia atari.jl --seed %o --log pooyan_%o.log --id pooyan
+430-439 julia atari.jl --seed %o --log private_eye_%o.log --id private_eye
+440-449 julia atari.jl --seed %o --log qbert_%o.log --id qbert
+450-459 julia atari.jl --seed %o --log riverraid_%o.log --id riverraid
+460-469 julia atari.jl --seed %o --log road_runner_%o.log --id road_runner
+470-479 julia atari.jl --seed %o --log pong_1%o.log --id pong
+480-489 julia atari.jl --seed %o --log robotank_%o.log --id robotank
+490-499 julia atari.jl --seed %o --log seaquest_%o.log --id seaquest
+500-509 julia atari.jl --seed %o --log skiing_%o.log --id skiing
+510-519 julia atari.jl --seed %o --log solaris_%o.log --id solaris
+520-529 julia atari.jl --seed %o --log space_invaders_%o.log --id space_invaders
+530-539 julia atari.jl --seed %o --log star_gunner_%o.log --id star_gunner
+540-549 julia atari.jl --seed %o --log tennis_%o.log --id tennis
+550-559 julia atari.jl --seed %o --log time_pilot_%o.log --id time_pilot
+560-569 julia atari.jl --seed %o --log tutankham_%o.log --id tutankham
+570-579 julia atari.jl --seed %o --log up_n_down_%o.log --id up_n_down
+580-589 julia atari.jl --seed %o --log venture_%o.log --id venture
+590-599 julia atari.jl --seed %o --log video_pinball_%o.log --id video_pinball
+600-609 julia atari.jl --seed %o --log wizard_of_wor_%o.log --id wizard_of_wor
+610-619 julia atari.jl --seed %o --log yars_revenge_%o.log --id yars_revenge
+620-629 julia atari.jl --seed %o --log zaxxon_%o.log --id zaxxon
+630-639 julia atari.jl --seed %o --log boxing_1%o.log --id boxing

scripts/irace.sh

@@ -1,7 +1,7 @@
-#!/bin/sh
+#!/bin/bash
 #SBATCH -J PCGP
-#SBATCH -N 10
-#SBATCH -n 200
+#SBATCH -N 5
+#SBATCH -n 100
 #SBATCH [email protected]
 #SBATCH --mail-type=ALL
 
@@ -15,4 +15,4 @@ cp -r $CGP/tuning/* $WORK_DIR/
 cp -r $CGP/experiments/atari.jl $WORK_DIR/
 cd $WORK_DIR
 
-irace --parallel 200 2>&1 > irace.log
+mpirun -np 1 irace --mpi 1 --parallel 99 2>&1 > irace.log

scripts/modload.sh

@@ -4,4 +4,5 @@ module purge
 module load intel/14.0.2.144
 module load julia
 module load R
+module load openmpi/3.0.0
 export PATH=/usr/local/julia/0.5.0_stand/julia-3c9d75391c/bin:$PATH