rm emb related

games/game.py

@@ -127,116 +127,3 @@ class game:
 
         return res
 
-
-    def _compute_emb(self, command: str):
-        # always normalize text when query OpenAI embedding service
-        command = command.strip().lower()
-        request_timeout = 5
-        return openai_get_embedding(
-            command,
-            engine="text-embedding-ada-002",
-            request_timeout=request_timeout,
-        )
-
-
-    def init_emb(self):
-        '''
-        initialize embeddings
-
-        1. load pkl file that is in train
-        2. run against OpenAI embedding service, but skip entires in pkl
-        '''
-
-        print('initializing...')
-        # load from the pkl file
-        pkl_embeddings = {}
-        try:
-            print('loading embeddings from a file...')
-            # file_path = f'{self.emb_path}'
-            # print(file_path)
-            # with open(file_path, 'rb') as f:
-            #     pkl_embeddings = pkl.load(f)
-            files = glob.glob(f"{self.ROOT}/data/{self.PATH}/emb/*.json")
-            pkl_embeddings = {}
-            for f in files:
-                with open(f, "r") as f:
-                    pkl_embeddings |= json.load(f)
-
-            # copy to self.embeddings, skip possible junk
-            for cmd in self.train:
-                if cmd in pkl_embeddings:
-                    self.embeddings[cmd] = pkl_embeddings[cmd]
-        except Exception as e:
-            print(e)
-        print(f'pkl size: {len(pkl_embeddings)}')
-        print(f'embedding size: {len(self.embeddings)}')
-
-        # query OpenAI embedding service
-        print('creating embeddings by query OpenAI...')
-        for cmd in self.train:
-            if cmd in self.embeddings:
-                # print(f' - skip {command}')
-                continue
-            print(f'query \'{cmd}\'...')
-            self.embeddings[cmd] = self._compute_emb(cmd)
-        # print(json.dumps(self.embeddings, indent=2))
-        print(f'embedding size: {len(self.embeddings)}')
-        print(f'train size: {len(self.train)}')
-        assert self.embeddings.keys() == self.train.keys()
-
-
-    def query_emb(self, command:str):
-        command = self.special_case(command)
-        
-        if len(self.embeddings) == 0:
-            # place init_emb here, not __init__,
-            # as we can avoid network connection when started.
-            # In this way, we can turn off VPN before starting,
-            # and turn on VPN after started.
-            self.init_emb()
-        else:
-            print('skip init')
-        if len(self.embeddings) == 0:
-            return self.status_failure, ['AI internal error', 'embeding is empty'], 'no desc', [['', 0]]
-                
-        if command is None:
-            return self.status_failure, ["Your input is empty."], 'no desc', [['', 0]]
-        if command.strip() == '':
-            return self.status_failure, ["Your input is empty."], 'no desc', [['', 0]]
-
-        imput_emb = self._compute_emb(command)
-
-        results = []
-        for cmd in self.embeddings:
-            sim = openai_cosine_similarity(
-                imput_emb, 
-                self.embeddings[cmd],
-            )
-            results.append([cmd, sim])
-        results.sort(key = lambda x: x[1], reverse=True)
-        # shorten results 
-        topk = 3 # 10
-        results = results[:topk]
-
-        top_cmd = None
-        top_sim = 0
-        if results:
-            top_cmd, top_sim = results[0]
-
-        if top_sim > 0.8:
-            return [
-                self.status_OK, 
-                self.train[top_cmd]["result"], 
-                self.train[top_cmd]['desc'],
-                results,
-            ]
-        else:
-            return [
-                self.status_failure, 
-                [
-                    "We don't understand your request.",
-                    f"Our best guess is: '{top_cmd}'"
-                ], 
-                'no desc',
-                results,
-            ]