X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=main.py;h=69731ff89e5e290b2124d84eeffad5aafcabef03;hb=9d8e9d4fac19329a328dd33b26115792a8090c57;hp=3ce5916aed336f38b3a66970c2d8af30639e2b5c;hpb=71eeffe7e9e3b379fe2c91e833dbeb35e1cb7971;p=picoclvr.git

diff --git a/main.py b/main.py
index 3ce5916..69731ff 100755
--- a/main.py
+++ b/main.py
@@ -5,7 +5,7 @@
 
 # Written by Francois Fleuret <francois@fleuret.org>
 
-import math, sys, argparse, time, tqdm, os
+import math, sys, argparse, time, tqdm, os, datetime
 
 import torch, torchvision
 from torch import nn
@@ -33,7 +33,7 @@ parser.add_argument(
     "--task",
     type=str,
     default="twotargets",
-    help="byheart, learnop, guessop, mixing, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl, grid, qmlp",
+    help="byheart, learnop, guessop, mixing, memory, twotargets, addition, picoclvr, mnist, maze, snake, stack, expr, rpl, grid, qmlp",
 )
 
 parser.add_argument("--log_filename", type=str, default="train.log", help=" ")
@@ -256,6 +256,12 @@ default_task_args = {
         "nb_train_samples": 50000,
         "nb_test_samples": 10000,
     },
+    "memory": {
+        "model": "37M",
+        "batch_size": 100,
+        "nb_train_samples": 25000,
+        "nb_test_samples": 1000,
+    },
     "mixing": {
         "model": "37M",
         "batch_size": 25,
@@ -285,6 +291,13 @@ default_model_args = {
         "nb_heads": 2,
         "nb_blocks": 2,
     },
+    "4M": {
+        "dim_model": 256,
+        "dim_keys": 32,
+        "dim_hidden": 1024,
+        "nb_heads": 4,
+        "nb_blocks": 6,
+    },
     "37M": {
         "dim_model": 512,
         "dim_keys": 64,
@@ -348,6 +361,8 @@ def log_string(s):
     sys.stdout.flush()
 
 
+log_string(f"argv {' '.join(sys.argv)}")
+
 for n in vars(args):
     log_string(f"args.{n} {getattr(args, n)}")
 
@@ -416,6 +431,16 @@ elif args.task == "twotargets":
         device=device,
     )
 
+elif args.task == "memory":
+    task = tasks.SandBox(
+        problem=problems.ProblemMemory(),
+        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 == "mixing":
     task = tasks.SandBox(
         problem=problems.ProblemMixing(
@@ -693,6 +718,8 @@ if nb_epochs_finished >= nb_epochs:
         deterministic_synthesis=args.deterministic_synthesis,
     )
 
+time_pred_result = None
+
 for n_epoch in range(nb_epochs_finished, nb_epochs):
     learning_rate = learning_rate_schedule[n_epoch]
 
@@ -751,6 +778,13 @@ for n_epoch in range(nb_epochs_finished, nb_epochs):
             deterministic_synthesis=args.deterministic_synthesis,
         )
 
+        time_current_result = datetime.datetime.now()
+        if time_pred_result is not None:
+            log_string(
+                f"next_result {time_current_result + (time_current_result - time_pred_result)}"
+            )
+        time_pred_result = time_current_result
+
     checkpoint = {
         "nb_epochs_finished": n_epoch + 1,
         "model_state": model.state_dict(),