X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=rpl.py;h=b848afa756a6188a60d29a446bdd51d1e455b74a;hb=HEAD;hp=7c1c96e2cd1f567f2c830c1d6040b79aaa436f47;hpb=5d46a9bd7d032d90ef4c4b38ac3c9b5b66526527;p=picoclvr.git

diff --git a/rpl.py b/rpl.py
index 7c1c96e..b848afa 100755
--- a/rpl.py
+++ b/rpl.py
@@ -58,7 +58,9 @@ rpl_ops = ["add", "min", "max", "swp", "rep", "dup", "del"]
 ######################################################################
 
 
-def generate(nb_starting_values=3, max_input=9, prog_len=6, nb_runs=5):
+def generate(
+    nb_starting_values=3, nb_result_values_max=None, max_input=9, prog_len=6, nb_runs=5
+):
     prog_len = (1 + torch.randint(2 * prog_len, (1,))).clamp(max=prog_len).item()
 
     while True:
@@ -73,11 +75,14 @@ def generate(nb_starting_values=3, max_input=9, prog_len=6, nb_runs=5):
             result_stack = rpl_exec(prog, stack)
             if len(result_stack) == 0:
                 no_empty_stack = False
-            result = result + ["<input>"] + stack + ["<output>"] + result_stack
+            result = result + ["<in>"] + stack + ["<out>"] + result_stack
 
-        result = result + ["<prog>"] + prog
+        result = result + ["<prg>"] + prog
         result = result + ["<end>"]
-        if no_empty_stack:
+
+        if no_empty_stack and (
+            nb_result_values_max is None or len(result_stack) <= nb_result_values_max
+        ):
             break
 
     return result
@@ -98,29 +103,44 @@ def next_marker(seq, tokens, start=0):
 def decompose(seq):
     io = []
     k = 0
-    while seq[k] == "<input>":
-        o = next_marker(seq, ["<output>"], start=k + 1)
-        e = next_marker(seq, ["<input>", "<prog>"], start=o)
-        if o is None or e is None:
-            raise ValueError("Invalid input/output")
+    while seq[k] == "<in>":
+        o = next_marker(seq, ["<out>"], start=k + 1)
+        if o is None:
+            raise ValueError("Missing output markers (should be correct in the prompt)")
+        e = next_marker(seq, ["<in>", "<prg>"], start=o)
+        if e is None:
+            raise ValueError(
+                "Missing input/output markers (should be correct in the prompt)"
+            )
         try:
             io.append(
                 ([int(x) for x in seq[k + 1 : o]], [int(x) for x in seq[o + 1 : e]])
             )
         except ValueError:
-            raise ValueError("Invalid input/output")
+            raise ValueError(
+                "Invalid input/output value (should be correct in the prompt)"
+            )
 
         k = e
 
-    if seq[k] == "<prog>":
+    if seq[k] == "<prg>":
         e = next_marker(seq, ["<end>"], start=k)
         if e is None:
             prog = []
         else:
             prog = seq[k + 1 : e]
+    else:
+        raise ValueError("Missing <prg> (it should be in the prompt)")
+
     return prog, io
 
 
+def stack_distance(target_stack, result_stack):
+    return abs(len(result_stack) - len(target_stack)) + sum(
+        [0 if x == y else 1 for x, y in zip(result_stack, target_stack)]
+    )
+
+
 def compute_nb_errors(seq):
     prog, io = decompose(seq)
 
@@ -140,9 +160,7 @@ def compute_nb_errors(seq):
         for start_stack, target_stack in io:
             result_stack = rpl_exec(prog, start_stack)
             nb_total += len(target_stack)
-            e = abs(len(result_stack) - len(target_stack)) + sum(
-                [0 if x == y else 1 for x, y in zip(result_stack, target_stack)]
-            )
+            e = stack_distance(target_stack, result_stack)
             nb_errors += e
             stacks.append((start_stack, target_stack, result_stack, e == 0))