CODEWITHSUNDEEP
pythontensorflowmachine-learningtensorflow-litetensorflow-hub
Ali Memon
Ali Memon

Reputation: 135

Save the Universal Sentence Encoder to Tflite or serve it to tensorflow api

I have this code for finding sentence similarity using the pre-built universal sentence encoder. It takes a .txt file as input. Performs cosine similarity and then accepts an output from user to find the most similar sentence as per users input query. This is the code:

# tensroflow hub module for Universal sentence Encoder
module_url = "https://tfhub.dev/google/universal-sentence-encoder-large/3" #@param ["https://tfhub.dev/google/universal-sentence-encoder/2", "https://tfhub.dev/google/universal-sentence-encoder-large/3"]

def get_features(texts):
    if type(texts) is str:
        texts = [texts]
    with tf.Session() as sess:
        sess.run([tf.global_variables_initializer(), tf.tables_initializer()])
        return sess.run(embed(texts))
    def remove_stopwords(stop_words, tokens):
    res = []
    for token in tokens:
        if not token in stop_words:
            res.append(token)
    return res

def process_text(text):
    text = text.encode('ascii', errors='ignore').decode()
    text = text.lower()
    text = re.sub(r'http\S+', ' ', text)
    text = re.sub(r'#+', ' ', text )
    text = re.sub(r'@[A-Za-z0-9]+', ' ', text)
    text = re.sub(r"([A-Za-z]+)'s", r"\1 is", text)
    #text = re.sub(r"\'s", " ", text)
    text = re.sub(r"\'ve", " have ", text)
    text = re.sub(r"won't", "will not ", text)
    text = re.sub(r"isn't", "is not ", text)
    text = re.sub(r"can't", "can not ", text)
    text = re.sub(r"n't", " not ", text)
    text = re.sub(r"i'm", "i am ", text)
    text = re.sub(r"\'re", " are ", text)
    text = re.sub(r"\'d", " would ", text)
    text = re.sub(r"\'ll", " will ", text)
    text = re.sub('\W', ' ', text)
    text = re.sub(r'\d+', ' ', text)
    text = re.sub('\s+', ' ', text)
    text = text.strip()
    return text

def lemmatize(tokens):
    lemmatizer = nltk.stem.WordNetLemmatizer()
    lemma_list = []
    for token in tokens:
        lemma = lemmatizer.lemmatize(token, 'v')
        if lemma == token:
            lemma = lemmatizer.lemmatize(token)
        lemma_list.append(lemma)
    # return [ lemmatizer.lemmatize(token, 'v') for token in tokens ]
    return lemma_list


def process_all(text):
    text = process_text(text)
    return ' '.join(remove_stopwords(stop_words, text.split()))

process_text("Hello! Who are you?")

with open('/content/sample_data/training.txt') as f:
...     text = [i.strip() for i in f]
...     

data_processed = list(map(process_text, text))
len(data_processed)

BASE_VECTORS = get_features(text)

def cosine_similarity(v1, v2):
    mag1 = np.linalg.norm(v1)
    mag2 = np.linalg.norm(v2)
    if (not mag1) or (not mag2):
        return 0
    return np.dot(v1, v2) / (mag1 * mag2)

def test_similiarity(text1, text2):
    vec1 = get_features(text1)[0]
    vec2 = get_features(text2)[0]
    print(vec1.shape)
    return cosine_similarity(vec1, vec2)

def semantic_search(query, data, vectors):
    query = process_text(query)
    print("Extracting features...")
    query_vec = get_features(query)[0].ravel()
    res = []
    for i, d in enumerate(data):
        qvec = vectors[i].ravel()
        sim = cosine_similarity(query_vec, qvec)
        res.append((sim, d[:100], i))
    return sorted(res, key=lambda x : x[0], reverse=True)

semantic_search("da vinci", data_processed, BASE_VECTORS)

I want to save the model and convert it to tflite. I have researched a lot but haven't been able to find any solution. Or how to serve it to tensorflow api.

Upvotes: 3

Views: 810

Answers (1)

Andrey Khorlin
Andrey Khorlin

Reputation: 213

One option to proceed would be to save the model in SavedModel format, then convert the resulting model to tflite. Note that the ability to convert the model may depend on the ops that the model is using and some model architectures may not be convertible to the tflite format.

Upvotes: 1

Related Questions