update

console_play.py

@@ -122,7 +122,7 @@ def play_cached_hospital():
 
 
 if __name__ == '__main__':
-    play_constrained()
+    play_unconstrained()
 
 
 

generator/ctrl/ctrl_generator.py

@@ -20,7 +20,7 @@ def loss(labels, logits):
 
 class CTRLGenerator():
 
-    def __init__(self, control_code="Writing Text: ", generate_num=40, temperature=0.3, topk=20, nucleus_prob=0):
+    def __init__(self, control_code="Writing Text: ", generate_num=32, temperature=0.3, topk=20, nucleus_prob=0):
 
         self.generate_num=generate_num
         model_dir = "generator/ctrl/training_utils/seqlen256_v1.ckpt/"
@@ -219,10 +219,7 @@ class CTRLGenerator():
         forbidden_tokens = ['<unk>', 'Sco@@', "&amp@@", "1]@@", "2]@@", "3]@@", "4]@@", "https://www.@@", "[@@", ":@@",
                             "Edit", "&@@", "2:","1:", ":", "Edit@@", "EDI@@", "EDIT@@", "edit", "TL@@", "tl@@", ";@@",
                             '**', "http://@@", "Redd@@", "UP@@", "mom", "Up@@", "Me:", "Update", "mom@@", "Part",
-                            "http://www.@@", "edit@@", "*@@", "\n", "Writing", "Text@@", "\\@@", "<br>@@", "<div", "|@@"]
-
-        if num_new_lines > self.max_new_lines:
-            forbidden_tokens.append("\n")
+                            "http://www.@@", "edit@@", "*@@", "Writing", "Text@@", "\\@@", "<br>@@", "<div", "|@@"]
 
         for forbidden_token in forbidden_tokens:
             prompt_logits[_token][self.word2idx[forbidden_token]] = -1e8
@@ -266,6 +263,8 @@ class CTRLGenerator():
         # assign the token for generation
         tokens_generated[0][token + 1] = idx
 
+        "\n",
+
         return idx
 
     def generate(self, prompt, options=None):
@@ -296,6 +295,8 @@ class CTRLGenerator():
             idx = self.generate_next_token(token, tokens_generated, options, num_new_lines, token_num, first_token=first_token)
             if self.idx2word[idx] is "\n":
                 num_new_lines += 1
+                if num_new_lines > 1:
+                    self.result_replace(result)
 
             print(repr(self.idx2word[idx]), end="_")
 

generator/ctrl/training_utils/make_forest_tf_records.py

@@ -20,7 +20,7 @@ def make_samples_helper(context, story_block, action_results, path, tree_id):
         new_path = path[:]
         new_path.append(i)
         if action_result["result"] is not None:
-            sample = [context, story_block, "action:", action_result["action"], "result:", action_result["result"]]
+            sample = [context, story_block, action_result["action"], action_result["result"]]
             samples.append(sample)
         if len(action_result["action_results"]) is not 0:
             sub_result = make_samples_helper(context, action_result["result"], action_result["action_results"], new_path, tree_id)
@@ -43,8 +43,8 @@ def build_tokenized_samples(bpe, tree):
     for sample in samples:
         sample = [string.strip() for string in sample]
 
-        sample[3] = sample[3][0].lower() + sample[3][1:]
-        sample[3] = "You " + sample[3]
+        sample[2] = sample[2][0].lower() + sample[2][1:]
+        sample[2] = "You " + sample[2]
         string_samples.append(" ".join(sample))
 
     tokenized_samples = [bpe.apply([sample.encode('ascii', errors='ignore') if not use_py3 else sample])[0] for sample in

story/story_manager.py

@@ -212,7 +212,7 @@ class CTRLStoryManager(ConstrainedStoryManager):
         results = []
         options = {"word_blacklist": {0:[]}}
         for phrase in self.action_phrases:
-            result = self.generate_action_result(self.story_context() + " action: ", phrase, options=options)
+            result = self.generate_action_result(self.story_context(), phrase, options=options)
             action_verb = result[0].split()[1]
             print("blacklisted verb is ", action_verb)
             options["word_blacklist"][0].append(action_verb)