Can't load dataframe columns by tf.data.Dataset.from_tensor_slices()

Question

I have a dataframe which consist of columns = id, Text, Media_location (which is relative path to images folder).

Now, I'm trying to load the columns Text, Media_location like this:

features = df[['Text', 'Media_location']]
dataset = tf.data.Dataset.from_tensor_slices((features))

and then this error comes up:

Exception has occurred: ValueError
Failed to convert a NumPy array to a Tensor (Unsupported object type float).

During handling of the above exception, another exception occurred:

  File "D:\Final\MultiCNN_test.py", line 114, in process_text_image
    dataset = tf.data.Dataset.from_tensor_slices((features))

I think this error is coming as the dataframe columns are not being able to convert to a tensor but I'm not sure how to do so, so as to remove the error.

Answer 1

If the columns Text and Media_location have the same data type your code will work:

import tensorflow as tf
import pandas as pd

df = pd.DataFrame(data={'Text': ['some text', 'some more text'],
                        'Media_location': ['/path/to/file1', '/path/to/file2']})

features = df[['Text', 'Media_location']]
dataset = tf.data.Dataset.from_tensor_slices((features))

for x in dataset:
  print(x)

tf.Tensor([b'some text' b'/path/to/file1'], shape=(2,), dtype=string)
tf.Tensor([b'some more text' b'/path/to/file2'], shape=(2,), dtype=string)

However, if both have different data types, you will get your error or a similar one, since a tensor cannot have mixed data types. So try something like this:

df = pd.DataFrame(data={'Text': [0.29, 0.58],
                        'Media_location': ['/path/to/file1', '/path/to/file2']})

dataset = tf.data.Dataset.from_tensor_slices((df['Text'], df['Media_location']))

for x in dataset:
  print(x)

(<tf.Tensor: shape=(), dtype=float64, numpy=0.29>, <tf.Tensor: shape=(), dtype=string, numpy=b'/path/to/file1'>)
(<tf.Tensor: shape=(), dtype=float64, numpy=0.58>, <tf.Tensor: shape=(), dtype=string, numpy=b'/path/to/file2'>)