X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=tasks.py;h=5019aed3b0953a7037bc213296ca7371d1b3c279;hb=6045e9a7dd61f0dab60bd1c6ff71f6bd5c32778b;hp=421aee49f2f0005f7650ea9836fde801fc5598e8;hpb=db7cefe4fefb381e56f1292d5bbe4a18c76afb47;p=picoclvr.git

diff --git a/tasks.py b/tasks.py
index 421aee4..5019aed 100755
--- a/tasks.py
+++ b/tasks.py
@@ -76,6 +76,7 @@ class Task:
 
 import problems
 
+
 class SandBox(Task):
     def __init__(
         self,
@@ -180,6 +181,43 @@ class SandBox(Task):
             f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
         )
 
+        logger(f"main_test_accuracy {n_epoch} {test_nb_correct/test_nb_total}")
+
+        if save_attention_image is None:
+            logger("no save_attention_image (is pycairo installed?)")
+        else:
+            for k in range(10):
+                ns = torch.randint(self.test_input.size(0), (1,)).item()
+                input = self.test_input[ns : ns + 1].clone()
+
+                with torch.autograd.no_grad():
+                    t = model.training
+                    model.eval()
+                    model.record_attention(True)
+                    model(BracketedSequence(input))
+                    model.train(t)
+                    ram = model.retrieve_attention()
+                    model.record_attention(False)
+
+                tokens_output = [c for c in self.problem.seq2str(input[0])]
+                tokens_input = ["n/a"] + tokens_output[:-1]
+                for n_head in range(ram[0].size(1)):
+                    filename = os.path.join(
+                        result_dir, f"sandbox_attention_{k}_h{n_head}.pdf"
+                    )
+                    attention_matrices = [m[0, n_head] for m in ram]
+                    save_attention_image(
+                        filename,
+                        tokens_input,
+                        tokens_output,
+                        attention_matrices,
+                        k_top=10,
+                        # min_total_attention=0.9,
+                        token_gap=12,
+                        layer_gap=50,
+                    )
+                    logger(f"wrote {filename}")
+
 
 ######################################################################
 
@@ -335,6 +373,10 @@ class PicoCLVR(Task):
             f"property_{prefix}miss {n_epoch} {100*nb_missing_properties/nb_requested_properties:.02f}%"
         )
 
+        logger(
+            f"main_test_accuracy {n_epoch} {1-nb_missing_properties/nb_requested_properties}"
+        )
+
     ######################################################################
 
     def produce_results(
@@ -605,6 +647,8 @@ class Maze(Task):
             f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
         )
 
+        logger(f"main_test_accuracy {n_epoch} {test_nb_correct/test_nb_total}")
+
         if count is not None:
             proportion_optimal = count.diagonal().sum().float() / count.sum()
             logger(f"proportion_optimal_test {proportion_optimal*100:.02f}%")
@@ -744,6 +788,8 @@ class Snake(Task):
             f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
         )
 
+        logger(f"main_test_accuracy {n_epoch} {test_nb_correct/test_nb_total}")
+
 
 ######################################################################
 
@@ -853,6 +899,8 @@ class Stack(Task):
             f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
         )
 
+        logger(f"main_test_accuracy {n_epoch} {test_nb_correct/test_nb_total}")
+
         ##############################################################
         # Log a few generated sequences
         input = self.test_input[:10, : 12 * (1 + self.nb_digits)]
@@ -1125,6 +1173,8 @@ class RPL(Task):
                 f"accuracy_prog_test {n_epoch} nb_total {test_nb_total} nb_errors {test_nb_errors} accuracy {100.0*(1-test_nb_errors/test_nb_total):.02f}%"
             )
 
+            logger(f"main_test_accuracy {n_epoch} {1-test_nb_errors/test_nb_total}")
+
         test_nb_total, test_nb_errors = compute_nb_errors_output(
             self.test_input[:1000].to(self.device), nb_to_log=10
         )
@@ -1133,9 +1183,11 @@ class RPL(Task):
             f"accuracy_output_test {n_epoch} nb_total {test_nb_total} nb_errors {test_nb_errors} accuracy {100.0*(1-test_nb_errors/test_nb_total):.02f}%"
         )
 
-        if save_attention_image is not None:
-            ns=torch.randint(self.test_input.size(0),(1,)).item()
-            input = self.test_input[ns:ns+1].clone()
+        if save_attention_image is None:
+            logger("no save_attention_image (is pycairo installed?)")
+        else:
+            ns = torch.randint(self.test_input.size(0), (1,)).item()
+            input = self.test_input[ns : ns + 1].clone()
             last = (input != self.t_nul).max(0).values.nonzero().max() + 3
             input = input[:, :last].to(self.device)
 
@@ -1321,6 +1373,8 @@ class Expr(Task):
             f"accuracy_test {n_epoch} nb_total {test_nb_total} nb_correct {test_nb_correct} accuracy {(100.0*test_nb_correct)/test_nb_total:.02f}%"
         )
 
+        logger(f"main_test_accuracy {n_epoch} {test_nb_correct/test_nb_total}")
+
         nb_total = test_nb_delta.sum() + test_nb_missed
         for d in range(test_nb_delta.size(0)):
             logger(