How To Convert Numpy Arrays To Standard Tensorflow Format?

I have two numpy arrays: One that contains captcha images Another that contains the corresponding labels (in one-hot vector format) I want to load these into TensorFlow so I can

Solution 1:

You can use tf.convert_to_tensor():

import tensorflow as tf
import numpy as np

data = [[1,2,3],[4,5,6]]
data_np = np.asarray(data, np.float32)

data_tf = tf.convert_to_tensor(data_np, np.float32)

sess = tf.InteractiveSession()  
print(data_tf.eval())

sess.close()

Here's a link to the documentation for this method:

https://www.tensorflow.org/api_docs/python/tf/convert_to_tensor

Solution 2:

You can use tf.pack (tf.stack in TensorFlow 1.0.0) method for this purpose. Here is how to pack a random image of type numpy.ndarray into a Tensor:

import numpy as np
import tensorflow as tf
random_image = np.random.randint(0,256, (300,400,3))
random_image_tensor = tf.pack(random_image)
tf.InteractiveSession()
evaluated_tensor = random_image_tensor.eval()

UPDATE: to convert a Python object to a Tensor you can use tf.convert_to_tensor function.

Solution 3:

You can use placeholders and feed_dict.

Suppose we have numpy arrays like these:

trX = np.linspace(-1, 1, 101) 
trY = 2 * trX + np.random.randn(*trX.shape) * 0.33

You can declare two placeholders:

X = tf.placeholder("float") 
Y = tf.placeholder("float")

Then, use these placeholders (X, and Y) in your model, cost, etc.: model = tf.mul(X, w) ... Y ... ...

Finally, when you run the model/cost, feed the numpy arrays using feed_dict:

with tf.Session() as sess:
.... 
    sess.run(model, feed_dict={X: trY, Y: trY})