TypeError：无法将 object 类型的 Sparsetensor 转换为 Tensor

Question

I am building a text classification model for imdb sentiment analysis dataset. I downloaded the dataset and followed the tutorial given here - https://developers.google.com/machine-learning/guides/text-classification/step-4

The error I get is

TypeError: Failed to convert object of type <class 'tensorflow.python.framework.sparse_tensor.SparseTensor'> to Tensor. Contents: SparseTensor(indices=Tensor("DeserializeSparse:0", shape=(None, 2), dtype=int64), values=Tensor("DeserializeSparse:1", shape=(None,), dtype=float32), dense_shape=Tensor("stack:0", shape=(2,), dtype=int64)). Consider casting elements to a supported type.

the type of x_train and x_val are scipy.sparse.csr.csr_matrix. This give an error when passed to sequential model. How to solve?

import tensorflow as tf
import numpy as np

from tensorflow.python.keras.preprocessing import sequence
from tensorflow.python.keras.preprocessing import text
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import f_classif

# Vectorization parameters

# Range (inclusive) of n-gram sizes for tokenizing text.
NGRAM_RANGE = (1, 2)

# Limit on the number of features. We use the top 20K features.
TOP_K = 20000

# Whether text should be split into word or character n-grams.
# One of 'word', 'char'.
TOKEN_MODE = 'word'

# Minimum document/corpus frequency below which a token will be discarded.
MIN_DOCUMENT_FREQUENCY = 2

# Limit on the length of text sequences. Sequences longer than this
# will be truncated.
MAX_SEQUENCE_LENGTH = 500


def ngram_vectorize(train_texts, train_labels, val_texts):
    """Vectorizes texts as ngram vectors.
    1 text = 1 tf-idf vector the length of vocabulary of uni-grams + bi-grams.
    # Arguments
        train_texts: list, training text strings.
        train_labels: np.ndarray, training labels.
        val_texts: list, validation text strings.
    # Returns
        x_train, x_val: vectorized training and validation texts
    """
    # Create keyword arguments to pass to the 'tf-idf' vectorizer.
    kwargs = {
            'ngram_range': NGRAM_RANGE,  # Use 1-grams + 2-grams.
            'dtype': 'int32',
            'strip_accents': 'unicode',
            'decode_error': 'replace',
            'analyzer': TOKEN_MODE,  # Split text into word tokens.
            'min_df': MIN_DOCUMENT_FREQUENCY,
    }
    vectorizer = TfidfVectorizer(**kwargs)

    # Learn vocabulary from training texts and vectorize training texts.
    x_train = vectorizer.fit_transform(train_texts)

    # Vectorize validation texts.
    x_val = vectorizer.transform(val_texts)

    # Select top 'k' of the vectorized features.
    selector = SelectKBest(f_classif, k=min(TOP_K, x_train.shape[1]))
    selector.fit(x_train, train_labels)
    x_train = selector.transform(x_train)
    x_val = selector.transform(x_val)

    x_train = x_train.astype('float32')
    x_val = x_val.astype('float32')
    return x_train, x_val

Answer 1

I also got the error message

TypeError: Failed to convert object of type <class 'tensorflow.python.framework.sparse_tensor.SparseTensor'> [...]

when I built a model based on the Google Machine Learning Guide for text classification .

Calling todense() on vectorized training and validation texts worked for me:

x_train = vectorizer.fit_transform(train_texts).todense()
x_val = vectorizer.transform(val_texts).todense()

(It seems to be very slow, though, I had to limit the training samples.)

EDIT:

When I remove this line (instead of adding .todense() ), it also seems to work:

model.add(Dropout(rate=dropout_rate, input_shape=x_train.shape[1:]))

For more details see this discussion: https://github.com/tensorflow/tensorflow/issues/47931

Answer 2

There's a similar open issue that you can find here .

Solution proposed is use Tensorflow version 2.1.0 and Keras version 2.3.1.

Answer 3

What I have done to solve this issue was convert the SpareTensors to arrays by calling .toarray()

So in your code, at the very end of your ngram_vectorize() function add:

return x_train.toarray(), x_val.toarray()