nltk 和 tensorflow 的问题需要无穷无尽的时间来预测

Question

如果我在我的 Mac 上运行，在最坏的情况下运行 model.predict() 需要 1 秒

但我想在服务器中部署机器人，所以我在 ubuntu 服务器（使用 vultr.com）中克隆了存储库并安装了 pip3 要求。

运行 python 脚本将阻止 model.predict() 中的执行，因此机器人无法正常工作

问题是脚本没有错误，但我认为是预测速度的问题。需要无限的时间

在 ubuntu 中使用 tensorflow 存在 cuda 和 tensorflow 兼容性问题，模型是用 cpu 而不是 gpu 预测的

但是是一个非常小的输入大小所以我想知道为什么 tensorflow predict 需要无限的时间

这是我的代码

from json import loads
from pickle import load

from nltk.stem import WordNetLemmatizer
from tensorflow.python.keras.models import load_model
from nltk import word_tokenize
from numpy import array

path = "/root/Jarvis_V3/process1_classification/my_nltk"

model = load_model(f'{path}/db/best_model.h5')
intents = loads(open(f'{path}/db/intents1.json').read())
words = load(open(f'{path}/db/words.pkl', 'rb'))
classes = load(open(f'{path}/db/classes.pkl', 'rb'))


class MyNltk:
    def __init__(self):
        self.lemmatizer = WordNetLemmatizer()


    def clean_up_sentence(self, sentence):

        # preprocessamento input utente
        print("tokenize")
        sentence_words = word_tokenize(sentence)
        print("lemmatizer")
        sentence_words = [self.lemmatizer.lemmatize(word.lower()) for word in sentence_words]
        print("end")
        return sentence_words

    def bow(self, sentence, words_, show_details=True):
        # creazione bag of words

        sentence_words = self.clean_up_sentence(sentence)
        print("bow")
        bag = [0] * len(words_)
        for s in sentence_words:
            for i, w in enumerate(words_):
                if w == s:
                    bag[i] = 1
                    if show_details:
                        print("found in bag: %s" % w)
        return array(bag)

    def calcola_pred(self, sentence, model_):

        p = self.bow(sentence, words, show_details=False)
        print("predction")
        **res = model_.predict(array([p]), verbose=1)[0]** # <-- HERE THE PROBLEM: INFINITE TIME EXECUTION
        print("end")
        ERROR_THRESHOLD = 0.25
        results = [[i, r] for i, r in enumerate(res) if r > ERROR_THRESHOLD]
        print("result")
        # sort by strength of probability
        results.sort(key=lambda x: x[1], reverse=True)
        return_list = []
        print(return_list)
        for r in results:
            return_list.append({"intent": classes[r[0]], "probability": str(r[1])})
        return return_list


    def getRisposta(self, ints, intents_json):
        # getting the possible targets
        tag = ints[0]['intent']
        list_of_intents = intents_json['intents']
        target = "none"
        for i in list_of_intents:
            if i['tag'] == tag:
                target = tag
                break
        return target

    def run(self, msg: str) -> str:
        print("nltk run")
        ints = self.calcola_pred(msg, model)
        target = self.getRisposta(ints, intents)
        return target

谢谢

我尝试解决了安装与tensorflow兼容的cuda依赖的问题

但是遇到很多问题比如

• 找不到cuda

• 安装失败无法安装cuda 11

• 如果安装了 cuda，tensorflow 无法看到正确的版本

• 我尝试删除、更新stackoverflow中与cuda相关的所有东西

比如

https://gist.github.com/ksopyla/bf74e8ce2683460d8de6e0dc389fc7f5 https://docs.vmware.com/en/VMware-vSphere-Bitfusion/3.0/Example-Guide/GUID-ABB4A0B1-F26E-422E-85C5-BA9F2454363A.html https://gist.github.com/ksopyla/bf74e8ce2683460d8de6e0dc389fc7f5 https://developer.nvidia.com/cuda-11.0-download-archive