X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=tasks.py;h=5019aed3b0953a7037bc213296ca7371d1b3c279;hb=960c93d7c0aea41d180814c46d3a05686a426764;hp=038a8ac55dc59d94a44787c542c23354446f22e1;hpb=6c8bed86221baae24a7c2aaaa41c009444efb5c9;p=picoclvr.git

diff --git a/tasks.py b/tasks.py
index 038a8ac..5019aed 100755
--- a/tasks.py
+++ b/tasks.py
@@ -181,37 +181,42 @@ 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}%"
         )
 
-        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()
-
-            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)
+        logger(f"main_test_accuracy {n_epoch} {test_nb_correct/test_nb_total}")
 
-            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"rpl_attention_{n_epoch}_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}")
+        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}")
 
 
 ######################################################################
@@ -368,6 +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(
@@ -638,6 +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}%")
@@ -777,6 +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}")
+
 
 ######################################################################
 
@@ -886,6 +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)]
@@ -1158,6 +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
         )
@@ -1166,7 +1183,9 @@ 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:
+        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
@@ -1354,6 +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(