CODEWITHSUNDEEP
pythontensorflow
David_tut
David_tut

Reputation: 43

What's the difference between tf.constant([1,2,3]) and tf.constant([[1,2,3]])

>>> tf.constant([1,2,3])
<tf.Tensor 'Const:0' shape=(3,) dtype=int32>
>>> tf.constant([[1,2,3]])
<tf.Tensor 'Const_1:0' shape=(1, 3) dtype=int32>

Is tf.constant([1,2,3]) creating a scalar and tf.constant([[1,2,3]]) creating an array?

Upvotes: 2

Views: 1307

Answers (1)

miradulo
miradulo

Reputation: 29720

Not quite. tf.constant([1, 2, 3]) creates a rank 1 constant tensor (a vector). Thus the shape is (3,).

>>> sess = tf.InteractiveSession()

>>> tf.constant([1, 2, 3]).eval()
array([1, 2, 3], dtype=int32)

While tf.constant([[1, 2, 3]]) creates a rank 2 constant tensor (a matrix), with 1 row and 3 columns.. so its shape is (1, 3). 

>>> tf.constant([[1, 2, 3]]).eval()
array([[1, 2, 3]], dtype=int32)

If you really wanted a scalar (rank 0), you wouldn't construct it with a sequence but rather just a scalar value. 

>>> tf.constant(3)
<tf.Tensor 'Const_5:0' shape=() dtype=int32>

Notice the empty shape here, making it clear it is rank 0.

See Tensor/Rank in the documentation.

Upvotes: 2

Related Questions