Update.

[picoclvr.git] / main.py

diff --git a/main.py b/main.py
index dbdf89d..704dff5 100755 (executable)
--- a/main.py
+++ b/main.py
@@ -33,7 +33,7 @@ parser.add_argument(
     "--task",
     type=str,
     default="twotargets",
-    help="byheart, learnop, guessop, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl",
+    help="byheart, learnop, guessop, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl, grid",
 )
 
 parser.add_argument("--log_filename", type=str, default="train.log", help=" ")
@@ -89,16 +89,21 @@ parser.add_argument("--checkpoint_name", type=str, default="checkpoint.pth")
 ##############################
 # rpl options
 
-parser.add_argument("--rpl_nb_starting_values", type=int, default=5)
+parser.add_argument("--rpl_nb_starting_values", type=int, default=3)
 
 parser.add_argument("--rpl_max_input", type=int, default=9)
 
-parser.add_argument("--rpl_prog_len", type=int, default=10)
+parser.add_argument("--rpl_prog_len", type=int, default=8)
 
-parser.add_argument("--rpl_nb_runs", type=int, default=8)
+parser.add_argument("--rpl_nb_runs", type=int, default=5)
 
 parser.add_argument("--rpl_no_prog", action="store_true", default=False)
 
+##############################
+# grid options
+
+parser.add_argument("--grid_size", type=int, default=6)
+
 ##############################
 # picoclvr options
 
@@ -113,11 +118,11 @@ parser.add_argument("--picocvlr_prune_properties", type=str, default="none")
 ##############################
 # Maze options
 
-parser.add_argument("--maze_height", type=int, default=23)
+parser.add_argument("--maze_height", type=int, default=13)
 
-parser.add_argument("--maze_width", type=int, default=39)
+parser.add_argument("--maze_width", type=int, default=21)
 
-parser.add_argument("--maze_nb_walls", type=int, default=45)
+parser.add_argument("--maze_nb_walls", type=int, default=15)
 
 ##############################
 # Snake options
@@ -173,37 +178,37 @@ if args.result_dir is None:
 default_task_args = {
     "byheart": {
         "model": "37M",
-        "nb_epochs": 5,
+        "nb_epochs": 2,
         "batch_size": 25,
         "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
     "learnop": {
         "model": "37M",
-        "nb_epochs": 5,
+        "nb_epochs": 15,
         "batch_size": 25,
         "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
     "guessop": {
-        "model": "122M",
+        "model": "352M",
         "nb_epochs": 5,
         "batch_size": 25,
-        "nb_train_samples": 250000,
+        "nb_train_samples": 1000000,
         "nb_test_samples": 10000,
     },
     "twotargets": {
         "model": "37M",
-        "nb_epochs": 5,
+        "nb_epochs": 10,
         "batch_size": 25,
         "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
     "addition": {
-        "model": "122M",
-        "nb_epochs": 5,
+        "model": "352M",
+        "nb_epochs": 50,
         "batch_size": 25,
-        "nb_train_samples": 50000,
+        "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
     "picoclvr": {
@@ -224,35 +229,35 @@ default_task_args = {
         "model": "37M",
         "nb_epochs": 25,
         "batch_size": 5,
-        "nb_train_samples": 250000,
+        "nb_train_samples": 100000,
         "nb_test_samples": 10000,
     },
     "snake": {
         "model": "37M",
         "nb_epochs": 5,
         "batch_size": 25,
-        "nb_train_samples": 50000,
+        "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
     "stack": {
         "model": "37M",
-        "nb_epochs": 5,
+        "nb_epochs": 15,
         "batch_size": 25,
         "nb_train_samples": 100000,
         "nb_test_samples": 1000,
     },
     "expr": {
-        "model": "37M",
-        "nb_epochs": 40,
+        "model": "352M",
+        "nb_epochs": 25,
         "batch_size": 25,
-        "nb_train_samples": 1000000,
+        "nb_train_samples": 2500000,
         "nb_test_samples": 10000,
     },
     "rpl": {
-        "model": "37M",
-        "nb_epochs": 40,
-        "batch_size": 25,
-        "nb_train_samples": 100000,
+        "model": "122M",
+        "nb_epochs": 50,
+        "batch_size": 5,
+        "nb_train_samples": 1000000,
         "nb_test_samples": 10000,
     },
     "world": {
@@ -262,6 +267,13 @@ default_task_args = {
         "nb_train_samples": 25000,
         "nb_test_samples": 1000,
     },
+    "grid": {
+        "model": "37M",
+        "nb_epochs": 25,
+        "batch_size": 25,
+        "nb_train_samples": 250000,
+        "nb_test_samples": 10000,
+    },
 }
 
 if args.task in default_task_args:
@@ -505,6 +517,16 @@ elif args.task == "rpl":
         device=device,
     )
 
+elif args.task == "grid":
+    task = tasks.Grid(
+        nb_train_samples=args.nb_train_samples,
+        nb_test_samples=args.nb_test_samples,
+        batch_size=args.batch_size,
+        size=args.grid_size,
+        logger=log_string,
+        device=device,
+    )
+
 elif args.task == "world":
     task = tasks.World(
         nb_train_samples=args.nb_train_samples,