fix: padding with new numpy versions

docs/source/text/index.rst

@@ -56,3 +56,4 @@ Adapters
 ********
 
 .. autoxpmconfig:: xpmir.text.adapters.MeanTextEncoder
+.. autoxpmconfig:: xpmir.text.adapters.TopicTextConverter

src/xpmir/neural/interaction/drmm.py

@@ -122,11 +122,17 @@ def _encode(
         options: TokenizerOptions,
     ) -> SimilarityInputWithTokens:
         encoded = encoder(texts, options=options)
+
+        max_len = max(encoded.tokenized.lens)
+        padded_tokens = [
+            (t + [""] * (max_len - len(t))) for t in encoded.tokenized.tokens
+        ]
+
         return self.similarity.preprocess(
             SimilarityInputWithTokens(
                 encoded.value,
                 encoded.tokenized.mask,
-                np.array(encoded.tokenized.tokens),
+                np.array(padded_tokens, dtype=str),
             )
         )
 

src/xpmir/utils/iter.py

@@ -406,13 +406,14 @@ def start(self):
 
     def close(self):
         if self.mp_iterator:
+            atexit.unregister(self.close)
             self.stop_process.set()
             try:
                 # Try to remove an item from the queue just in case
                 next(self.mp_iterator)
             finally:
+                self.mp_iterator = None
                 logging.info("Signaled the mp_iterator to quit")
-        atexit.unregister(self.close)
 
     def detach(self):
         """Produces an iterator only based on the multiprocess queue (useful