X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=main.py;h=f4e4f5c2c11e27d9be2666b81fc82d6e02f09b40;hb=8d9cd6a2c09da2105ca17b04df94fcf84e8de954;hp=303abc1bdc11c825ae3844ea37554a3cc9a157d2;hpb=3f09462033feac19ad72ac1a4b8690e6330df22d;p=picoclvr.git

diff --git a/main.py b/main.py
index 303abc1..f4e4f5c 100755
--- a/main.py
+++ b/main.py
@@ -32,8 +32,8 @@ parser = argparse.ArgumentParser(
 parser.add_argument(
     "--task",
     type=str,
-    default="sandbox",
-    help="byheart, learnop, guessop, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl",
+    default="twotargets",
+    help="byheart, learnop, guessop, mixing, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl, grid, qmlp",
 )
 
 parser.add_argument("--log_filename", type=str, default="train.log", help=" ")
@@ -46,7 +46,7 @@ parser.add_argument("--max_percents_of_test_in_train", type=int, default=1)
 
 ########################################
 
-parser.add_argument("--nb_epochs", type=int, default=None)
+parser.add_argument("--nb_epochs", type=int, default=25)
 
 parser.add_argument("--batch_size", type=int, default=None)
 
@@ -62,7 +62,7 @@ parser.add_argument("--learning_rate_schedule", type=str, default="10: 2e-5,30:
 
 ########################################
 
-parser.add_argument("--model", type=str, default="37M")
+parser.add_argument("--model", type=str, default=None)
 
 parser.add_argument("--dim_model", type=int, default=None)
 
@@ -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
@@ -155,9 +160,11 @@ parser.add_argument("--expr_result_max", type=int, default=99)
 parser.add_argument("--expr_input_file", type=str, default=None)
 
 ##############################
-# World options
+# Misc
 
-parser.add_argument("--world_vqae_nb_epochs", type=int, default=25)
+parser.add_argument("--mixing_hard", action="store_true", default=False)
+
+parser.add_argument("--mixing_deterministic_start", action="store_true", default=False)
 
 ######################################################################
 
@@ -171,83 +178,95 @@ if args.result_dir is None:
 ######################################################################
 
 default_task_args = {
+    "addition": {
+        "model": "352M",
+        "batch_size": 25,
+        "nb_train_samples": 250000,
+        "nb_test_samples": 10000,
+    },
     "byheart": {
-        "nb_epochs": 5,
+        "model": "37M",
         "batch_size": 25,
         "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
-    "learnop": {
-        "nb_epochs": 5,
+    "expr": {
+        "model": "352M",
         "batch_size": 25,
-        "nb_train_samples": 50000,
+        "nb_train_samples": 2500000,
         "nb_test_samples": 10000,
     },
-    "guessop": {
-        "nb_epochs": 5,
+    "grid": {
+        "model": "37M",
         "batch_size": 25,
-        "nb_train_samples": 50000,
+        "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
-    "twotargets": {
-        "nb_epochs": 5,
+    "qmlp": {
+        "model": "37M",
+        "batch_size": 10,
+        "nb_train_samples": 100000,
+        "nb_test_samples": 1000,
+    },
+    "guessop": {
+        "model": "352M",
         "batch_size": 25,
-        "nb_train_samples": 50000,
+        "nb_train_samples": 1000000,
         "nb_test_samples": 10000,
     },
-    "addition": {
-        "nb_epochs": 5,
+    "learnop": {
+        "model": "37M",
         "batch_size": 25,
         "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
+    "maze": {
+        "model": "37M",
+        "batch_size": 5,
+        "nb_train_samples": 100000,
+        "nb_test_samples": 10000,
+    },
     "picoclvr": {
-        "nb_epochs": 25,
+        "model": "37M",
         "batch_size": 25,
         "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
-    "mnist": {
-        "nb_epochs": 25,
-        "batch_size": 10,
-        "nb_train_samples": 60000,
-        "nb_test_samples": 10000,
-    },
-    "maze": {
-        "nb_epochs": 25,
+    "rpl": {
+        "model": "352M",
         "batch_size": 5,
-        "nb_train_samples": 250000,
+        "nb_train_samples": 2500000,
         "nb_test_samples": 10000,
     },
     "snake": {
-        "nb_epochs": 5,
+        "model": "37M",
         "batch_size": 25,
-        "nb_train_samples": 50000,
+        "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
     "stack": {
-        "nb_epochs": 5,
+        "model": "37M",
         "batch_size": 25,
         "nb_train_samples": 100000,
         "nb_test_samples": 1000,
     },
-    "expr": {
-        "nb_epochs": 40,
+    "twotargets": {
+        "model": "37M",
         "batch_size": 25,
-        "nb_train_samples": 1000000,
+        "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
-    "rpl": {
-        "nb_epochs": 40,
+    "mixing": {
+        "model": "37M",
         "batch_size": 25,
-        "nb_train_samples": 100000,
+        "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
-    "world": {
-        "nb_epochs": 10,
-        "batch_size": 25,
-        "nb_train_samples": 25000,
-        "nb_test_samples": 1000,
+    "mnist": {
+        "model": "37M",
+        "batch_size": 10,
+        "nb_train_samples": 60000,
+        "nb_test_samples": 10000,
     },
 }
 
@@ -397,6 +416,18 @@ elif args.task == "twotargets":
         device=device,
     )
 
+elif args.task == "mixing":
+    task = tasks.SandBox(
+        problem=problems.ProblemMixing(
+            hard=args.mixing_hard, random_start=not args.mixing_deterministic_start
+        ),
+        nb_train_samples=args.nb_train_samples,
+        nb_test_samples=args.nb_test_samples,
+        batch_size=args.batch_size,
+        logger=log_string,
+        device=device,
+    )
+
 elif args.task == "addition":
     task = tasks.SandBox(
         problem=problems.ProblemAddition(),
@@ -492,12 +523,22 @@ elif args.task == "rpl":
         device=device,
     )
 
-elif args.task == "world":
-    task = tasks.World(
+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,
-        vqae_nb_epochs=args.world_vqae_nb_epochs,
+        size=args.grid_size,
+        logger=log_string,
+        device=device,
+    )
+
+elif args.task == "qmlp":
+    task = tasks.QMLP(
+        nb_train_samples=args.nb_train_samples,
+        nb_test_samples=args.nb_test_samples,
+        batch_size=args.batch_size,
+        result_dir=args.result_dir,
         logger=log_string,
         device=device,
     )