X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=main.py;h=704dff5b95a918637e8cfc0282796322cb706fea;hb=6681907dcc86bf6e159925814d419f522e0e3300;hp=9c28e4789d86162f0be41763b2e27232d002ddd5;hpb=687d5b2d9f465577665991b84faec7c789685271;p=picoclvr.git

diff --git a/main.py b/main.py
index 9c28e47..704dff5 100755
--- a/main.py
+++ b/main.py
@@ -32,8 +32,8 @@ parser = argparse.ArgumentParser(
 parser.add_argument(
     "--task",
     type=str,
-    default="sandbox",
-    help="sandbox, picoclvr, mnist, maze, snake, stack, expr, rpl, world",
+    default="twotargets",
+    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=" ")
@@ -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)
 
@@ -89,26 +89,20 @@ 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)
 
 ##############################
-# sandbox options
+# grid options
 
-parser.add_argument("--sandbox_level", type=int, default=0)
-
-parser.add_argument("--sandbox_levels_nb_items", type=int, default=25)
-
-parser.add_argument("--sandbox_levels_len_source", type=int, default=6)
-
-parser.add_argument("--sandbox_levels_len_result", type=int, default=8)
+parser.add_argument("--grid_size", type=int, default=6)
 
 ##############################
 # picoclvr options
@@ -124,18 +118,18 @@ 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
 
-parser.add_argument("--snake_height", type=int, default=6)
+parser.add_argument("--snake_height", type=int, default=9)
 
-parser.add_argument("--snake_width", type=int, default=8)
+parser.add_argument("--snake_width", type=int, default=12)
 
 parser.add_argument("--snake_nb_colors", type=int, default=5)
 
@@ -182,60 +176,104 @@ if args.result_dir is None:
 ######################################################################
 
 default_task_args = {
-    "sandbox": {
+    "byheart": {
+        "model": "37M",
+        "nb_epochs": 2,
+        "batch_size": 25,
+        "nb_train_samples": 50000,
+        "nb_test_samples": 10000,
+    },
+    "learnop": {
+        "model": "37M",
+        "nb_epochs": 15,
+        "batch_size": 25,
+        "nb_train_samples": 50000,
+        "nb_test_samples": 10000,
+    },
+    "guessop": {
+        "model": "352M",
+        "nb_epochs": 5,
+        "batch_size": 25,
+        "nb_train_samples": 1000000,
+        "nb_test_samples": 10000,
+    },
+    "twotargets": {
+        "model": "37M",
+        "nb_epochs": 10,
+        "batch_size": 25,
+        "nb_train_samples": 50000,
+        "nb_test_samples": 10000,
+    },
+    "addition": {
+        "model": "352M",
         "nb_epochs": 50,
         "batch_size": 25,
-        "nb_train_samples": 100000,
+        "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
     "picoclvr": {
+        "model": "37M",
         "nb_epochs": 25,
         "batch_size": 25,
         "nb_train_samples": 250000,
         "nb_test_samples": 10000,
     },
     "mnist": {
+        "model": "37M",
         "nb_epochs": 25,
         "batch_size": 10,
-        "nb_train_samples": 250000,
+        "nb_train_samples": 60000,
         "nb_test_samples": 10000,
     },
     "maze": {
+        "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": 250000,
         "nb_test_samples": 10000,
     },
     "stack": {
-        "nb_epochs": 5,
+        "model": "37M",
+        "nb_epochs": 15,
         "batch_size": 25,
         "nb_train_samples": 100000,
         "nb_test_samples": 1000,
     },
     "expr": {
-        "nb_epochs": 40,
+        "model": "352M",
+        "nb_epochs": 25,
         "batch_size": 25,
-        "nb_train_samples": 1000000,
+        "nb_train_samples": 2500000,
         "nb_test_samples": 10000,
     },
     "rpl": {
-        "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": {
+        "model": "37M",
         "nb_epochs": 10,
         "batch_size": 25,
         "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:
@@ -341,32 +379,52 @@ picoclvr_pruner_eval = (
 
 ######################################################################
 
-if args.task == "sandbox":
-    if args.sandbox_level == 0:
-        problem = problems.ProblemLevel0(
-            nb_sentences=args.sandbox_levels_nb_items,
-            len_prompt=args.sandbox_levels_len_source,
-            len_result=args.sandbox_levels_len_result,
-        )
-    elif args.sandbox_level == 1:
-        problem = problems.ProblemLevel1(
-            nb_operators=args.sandbox_levels_nb_items,
-            len_source=args.sandbox_levels_len_source,
-            len_result=args.sandbox_levels_len_result,
-        )
-    elif args.sandbox_level == 2:
-        problem = problems.ProblemLevel2(
-            len_source=args.sandbox_levels_len_source,
-            len_result=args.sandbox_levels_len_result,
-        )
-    else:
-        raise ValueError(f"Unknown sandbox level {args.sandbox_level}")
+if args.task == "byheart":
+    task = tasks.SandBox(
+        problem=problems.ProblemByHeart(),
+        nb_train_samples=args.nb_train_samples,
+        nb_test_samples=args.nb_test_samples,
+        batch_size=args.batch_size,
+        logger=log_string,
+        device=device,
+    )
+    args.max_percents_of_test_in_train = -1
+
+elif args.task == "learnop":
+    task = tasks.SandBox(
+        problem=problems.ProblemLearnOperator(),
+        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 == "guessop":
+    task = tasks.SandBox(
+        problem=problems.ProblemGuessOperator(),
+        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 == "twotargets":
     task = tasks.SandBox(
-        # problem,
-        # problems.ProblemAddition(zero_padded=False, inverted_result=False),
-        # problems.ProblemLenId(len_max=args.sandbox_levels_len_source),
-        problems.ProblemTwoTargets(len_total=12, len_targets=4),
+        problem=problems.ProblemTwoTargets(),
+        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(),
         nb_train_samples=args.nb_train_samples,
         nb_test_samples=args.nb_test_samples,
         batch_size=args.batch_size,
@@ -459,6 +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,
@@ -554,33 +622,33 @@ train_set_perplexity = math.exp(entropy)
 ######################################################################
 # A bit of paranoia never hurts
 
+if args.max_percents_of_test_in_train >= 0:
+
+    def subsets_as_tuples(batches, cs):
+        s = set()
+        for batch in batches:
+            for x in batch:
+                s.add(tuple([v.item() for v in x]))
+                if len(s) == cs:
+                    yield s
+                    s = set()
+        yield s
+
+    nb_test, nb_in_train = 0, 0
+    for test_subset in subsets_as_tuples(task.batches(split="test"), 25000):
+        in_train = set()
+        for train_subset in subsets_as_tuples(task.batches(split="train"), 25000):
+            in_train.update(test_subset.intersection(train_subset))
+        nb_in_train += len(in_train)
+        nb_test += len(test_subset)
+
+    log_string(
+        f"data_check {nb_in_train*100/nb_test:.02f}% ({nb_in_train}/{nb_test}) of test samples are in the train set"
+    )
 
-def subsets_as_tuples(batches, cs):
-    s = set()
-    for batch in batches:
-        for x in batch:
-            s.add(tuple([v.item() for v in x]))
-            if len(s) == cs:
-                yield s
-                s = set()
-    yield s
-
-
-nb_test, nb_in_train = 0, 0
-for test_subset in subsets_as_tuples(task.batches(split="test"), 25000):
-    in_train = set()
-    for train_subset in subsets_as_tuples(task.batches(split="train"), 25000):
-        in_train.update(test_subset.intersection(train_subset))
-    nb_in_train += len(in_train)
-    nb_test += len(test_subset)
-
-log_string(
-    f"data_check {nb_in_train*100/nb_test:.02f}% ({nb_in_train}/{nb_test}) of test samples are in the train set"
-)
-
-assert (
-    nb_in_train <= args.max_percents_of_test_in_train * nb_test / 100
-), f"More than {args.max_percents_of_test_in_train}% of test samples are in the train set"
+    assert (
+        nb_in_train <= args.max_percents_of_test_in_train * nb_test / 100
+    ), f"More than {args.max_percents_of_test_in_train}% of test samples are in the train set"
 
 ##############################