hddlmZddlZddlmZddlmZmZmZm Z m Z m Z ddl m Z erddlZej eZe eeeeeefZddZddZGdd e d ZGd d Zy)) annotationsN)Sequence) TYPE_CHECKINGAnyCallableLiteralOptionalUnion) TypedDictcddl}t|dk(st|dk(r|jgS|j|}|j|}|jd|jdk7r&t d|jd|jd ddl}|j||j }|j||j }d|j||d z }t|tr|j|gS|j|S#t$rtjd |jj|d }|jj|d }|jd d 5|j ||j"|j$||z }dddn #1swYnxYwd|j&||j(|z<|cYSwxYw)z>? s2A9D)D))A,H 2G H G 1H H cR ddlmdfd }|S#t$r tdwxYw)zGet the OpenAI GPT-3 encoder.r)ClientzTHe default encoder for the EmbeddingDistance class uses the OpenAI API. Please either install the openai library with `pip install openai` or provide a custom encoder function (Callable[[str], Sequence[float]]).c}|jjt|d}|jDcgc]}|jc}Scc}w)Nztext-embedding-3-small)inputmodel) embeddingscreatelistdata embedding)textsclientresponsed OpenAIClients r3 encode_textz(_get_openai_encoder..encode_textIsM$$++u+%=, &.]]3 333sA)r?z Sequence[str]returnzSequence[Sequence[float]])openair6r )rDrCs @r3_get_openai_encoderrG>s< 14    T   s&c"eZdZUded<ded<y)EmbeddingConfigz0Callable[[list[str]], Sequence[Sequence[float]]]encoderzCLiteral['cosine', 'euclidean', 'manhattan', 'chebyshev', 'hamming']rN)__name__ __module__ __qualname____annotations__r3rIrISs == OOrPrIF)totalceZdZ d d dZ d dZd dZeddZed dZed dZ ed dZ ed d Z y)EmbeddingDistanceNc|xsi}|jdxsd|_|jdxs t|_y)NrrrJ)getdistancerGrJ)selfconfigs r3__init__zEmbeddingDistance.__init__Ys;2 8,8 zz),E0C0E rPc ddl}|j||g}|j|}|j |d|dj S#t$r tdwxYw)NrzWThe EmbeddingDistance class requires NumPy. Please install it with `pip install numpy`.r )rr rJr_compute_distanceitem)rW prediction referencer-r:vectors r3evaluatezEmbeddingDistance.evaluateasv    \\:y"9: *%%%fQi;@@BB '  s AA&c||jdk(r|j||S|jdk(r|j||S|jdk(r|j||S|jdk(r|j ||S|jdk(r|j ||St d|j)Nr euclidean manhattan chebyshevhammingzInvalid distance metric: )rV_cosine_distance_euclidean_distance_manhattan_distance_chebyshev_distance_hamming_distancer)rWabs r3r[z#EmbeddingDistance._compute_distanceqs ==H $((A. . ]]k )++Aq1 1 ]]k )++Aq1 1 ]]k )++Aq1 1 ]]i '))!Q/ /8HI IrPc$dt|g|gz S)zCompute the cosine distance between two vectors. Args: a (np.ndarray): The first vector. b (np.ndarray): The second vector. Returns: np.ndarray: The cosine distance. g?)r4rkrls r3rfz"EmbeddingDistance._cosine_distances&sQC000rPcFtjj||z S)zCompute the Euclidean distance between two vectors. Args: a (np.ndarray): The first vector. b (np.ndarray): The second vector. Returns: np.floating: The Euclidean distance. )r-r#r$rns r3rgz%EmbeddingDistance._euclidean_distancesyy~~a!e$$rPcXtjtj||z S)zCompute the Manhattan distance between two vectors. Args: a (np.ndarray): The first vector. b (np.ndarray): The second vector. Returns: np.floating: The Manhattan distance. )r-sumabsrns r3rhz%EmbeddingDistance._manhattan_distancevvbffQUm$$rPcXtjtj||z S)zCompute the Chebyshev distance between two vectors. Args: a (np.ndarray): The first vector. b (np.ndarray): The second vector. Returns: np.floating: The Chebyshev distance. )r-maxrrrns r3riz%EmbeddingDistance._chebyshev_distancersrPc2tj||k7S)zCompute the Hamming distance between two vectors. Args: a (np.ndarray): The first vector. b (np.ndarray): The second vector. Returns: np.floating: The Hamming distance. )r-meanrns r3rjz#EmbeddingDistance._hamming_distanceswwqAvrP)N)rXzOptional[EmbeddingConfig])r]strr^rxrEr)rk np.ndarrayrlryrEz np.floating)rkryrlryrEry) rKrLrMrYr`r[ staticmethodrfrgrhrirjrOrPr3rSrSXs-1F)FCCC  C J 1 1 % % % % % %  rPrS)r+Matrixr,r{rEry)rEz4Callable[[Sequence[str]], Sequence[Sequence[float]]]) __future__rloggingcollections.abcrtypingrrrrr r typing_extensionsr rr- getLoggerrKr!r<rr{r4rGrIrSrOrPr3rs|"$(   8 $ tDK $s)S0 1"J*PiuP ffrP