Question

我有一种用于文本分类的算法，可以正常工作。但是，它在预测中仅返回我一项。我想知道如何或是否有可能返回，每个新条目有3个项目。

火车：

def train(texts, marcas):
    print("Training.....")
    tvect = TfidfTransformer()  # -> Obsolete <- TfidfVectorizer(min_df=1, max_df=2, lowercase=False, preprocessor=None)
    # gera a vetorização (posição das palavras) para treino e predição
    count_vect = CountVectorizer(stop_words=None)
    vect_count_text = count_vect.fit_transform(texts)  # [vectorize_text(texto, tradutor) for texto in texts]
    vetoresDeTexto = tvect.fit_transform(vect_count_text)
    # Define o conjunto de dados X
    X = vetoresDeTexto  # np.array(vetoresDeTexto)
    # Define o conjunto de dados Y (labels)
    Y = np.array(marcas.tolist())
    # Define porcentagem do treino
    porcentagem_de_treino = 0.8
    # Separa o tamanho do treino a partir da porcentagem
    tamanho_do_treino = int(porcentagem_de_treino * len(Y))
    # O restante fica para a validacao
    tamanho_de_validacao = (len(Y) - tamanho_do_treino)
    # Separa os dados de treino
    treino_dados = X[0:tamanho_do_treino]
    # Separa as marcacoes de treino
    treino_marcacoes = Y[0:tamanho_do_treino]
    # Separa os dados de validacao
    validacao_dados = X[tamanho_do_treino:]
    # Separa as marcacoes de validacao
    validacao_marcacoes = Y[tamanho_do_treino:]
    print("Validacao Marcacoes: ")
    print(validacao_marcacoes)
    clf = LogisticRegression(class_weight=None)  # MultinomialNB() obtive 62% de acerto#GaussianNB()
    clf.fit(treino_dados, treino_marcacoes)
    # accuracy
    accuracy = clf.score(validacao_dados, validacao_marcacoes)
    file_name = 'train_data.pkl'
    cPickle.dump(clf, open(file_name, 'wb'))
    # fit_file = joblib.dump(clf, file_name)
    print("Accuracy: ")
    print("%.2f " % round(accuracy * 100) + "%\n")
    print("End of train...")
    predict(file_name, tvect, count_vect, treino_marcacoes, clf, treino_dados)
    # To get a fit_file
    # return fit_file

我的预测代码：

# to predict
def predict(fit, tvect, count_vect, y_test, clf, treinoX):
    print("\nPredict......")
    # new text to predict
    newTextToPredict = ["Just a new text to predict"]  # returns label J44

    new = count_vect.transform(newTextToPredict)
    # carrega o modelo treinado
    loaded_model = cPickle.load(open(fit, 'rb'))
    # faz a predição do novo texto de entrada
    result = loaded_model.predict(new)
    probs = clf.predict_proba(new)
    # precision_score(result, treinoY, average='samples')
    print(result)

当前输出：

....
End of train...

Predict......
['Z000']

此输出是否可能向我显示3个最可能的结果？

编辑：

我尝试使用predict_proba，但我不理解结果，它遵循输出：

print(probs)

Predict......
['Z000']
[[0.00472141 0.00468681 0.00545111 0.00473597 0.00742972 0.00459905
  0.00472848 0.00471651 0.00830986 0.00472729 0.00537823 0.00539556
  0.00463566 0.00469166 0.00473597 0.00469889 0.00473122 0.00510944
  0.00475248 0.00475248 0.00472681 0.00465737 0.0046238  0.00538928
  0.0053852  0.00469701 0.00470745 0.0052977  0.00468655 0.00472517
  0.00601271 0.00540062 0.00471387 0.00471311 0.00471592 0.00468392
  0.00470526 0.00454069 0.00467939 0.00471795 0.00706113 0.00475248
  0.00470356 0.00451991 0.00473597 0.02389303 0.00472151 0.00475248
  0.00573423 0.00469125 0.00471707 0.00450935 0.00458729 0.00607249
  0.00556578 0.00661622 0.00747174 0.00528275 0.00469896 0.00527276
  0.00537725 0.0046918  0.00472592 0.00523041 0.00466061 0.00523704
  0.00535152 0.00471286 0.00456425 0.00473597 0.00466597 0.00475248
  0.00471198 0.00470039 0.00545111 0.00473597 0.0059082  0.00471645
  0.0050765  0.00536772 0.00469146 0.0047054  0.00583113 0.00556937
  0.00530836 0.00724415 0.00499861 0.00469217 0.00471454 0.00456743
  0.00473241 0.00468181 0.00545604 0.00471984 0.00466745 0.00606397
  0.01230014 0.00467241 0.00472609 0.00541621 0.00473499 0.00468064
  0.00472712 0.00470356 0.00497979 0.00453495 0.00469214 0.00668041
  0.00528025 0.00468329 0.00777699 0.00468618 0.00537916 0.00455798
  0.0046802  0.00468039 0.00534045 0.00466915 0.00521349 0.00465117
  0.00466947 0.00688886 0.00460614 0.00648024 0.00469368 0.00456555
  0.2215044  0.01841092 0.00594679 0.00467938 0.01121442 0.00537937
  0.00468134 0.00472712 0.00470844 0.00470639 0.00580538 0.00535144
  0.00473597 0.00465237 0.00577107 0.00539569 0.00472306 0.00538426
  0.00472506]]

Answer 1

尝试打印probs：

probs = clf.predict_proba(new)
print(probs)

您将获得一系列概率。这些概率的总和等于 1 。然后，您可以从此数组中选择3个最高值的 values 元素索引。

top3_classes = np.argsort(probs)[:3]

这些是您需要的前3类标签索引。因此，您可以这样做：

print(your_classes[top3_classes])

您获得了预测的前三类。

Answer 2

更改您看待问题的方式。您想要的不是每个类“给我最可能的类”，而是“数据是否属于该类？”。如此多的二进制分类，而不是对具有多个类的单个分类进行澄清

（或者，还不清楚，您可能想获得可能性而不是argmax。在sklearn中，它是“ predict_proba”）

如何在预测中返回多个项目？

2 个答案: