Theano - Data Types


Advertisements

Now, that you have understood the basics of Theano, let us begin with the different data types available to you for creating your expressions. The following table gives you a partial list of data types defined in Theano.

Data type Theano type
Byte

bscalar, bvector, bmatrix, brow, bcol, btensor3, btensor4, btensor5, btensor6, btensor7

16-bit integers

wscalar, wvector, wmatrix, wrow, wcol, wtensor3, wtensor4, wtensor5, wtensor6, wtensor7

32-bit integers

iscalar, ivector, imatrix, irow, icol, itensor3, itensor4, itensor5, itensor6, itensor7

64-bit integers

lscalar, lvector, lmatrix, lrow, lcol, ltensor3, ltensor4, ltensor5, ltensor6, ltensor7

float

fscalar, fvector, fmatrix, frow, fcol, ftensor3, ftensor4, ftensor5, ftensor6, ftensor7

double

dscalar, dvector, dmatrix, drow, dcol, dtensor3, dtensor4, dtensor5, dtensor6, dtensor7

complex

cscalar, cvector, cmatrix, crow, ccol, ctensor3, ctensor4, ctensor5, ctensor6, ctensor7

The above list is not exhaustive and the reader is referred to the tensor creation document for a complete list.

I will now give you a few examples of how to create variables of various kinds of data in Theano.

Scalar

To construct a scalar variable you would use the syntax −

Syntax

x = theano.tensor.scalar ('x')
x = 5.0
print (x)

Output

5.0

One-dimensional Array

To create a one dimensional array, use the following declaration −

Example

f = theano.tensor.vector
f = (2.0, 5.0, 3.0)
print (f)f = theano.tensor.vector
f = (2.0, 5.0, 3.0)
print (f)
print (f[0])
print (f[2])

Output

(2.0, 5.0, 3.0)
2.0
3.0

If you do f[3] it would generate an index out of range error as shown here −

print f([3])

Output

IndexError                          Traceback (most recent call last)
<ipython-input-13-2a9c2a643c3a> in <module>
   4 print (f[0])
   5 print (f[2])
----> 6 print (f[3])
IndexError: tuple index out of range

Two-dimensional Array

To declare a two-dimensional array you would use the following code snippet −

Example

m = theano.tensor.matrix
m = ([2,3], [4,5], [2,4])
print (m[0])
print (m[1][0])

Output

[2, 3]
4

5-dimensional Array

To declare a 5-dimensional array, use the following syntax −

Example

m5 = theano.tensor.tensor5
m5 = ([0,1,2,3,4], [5,6,7,8,9], [10,11,12,13,14])
print (m5[1])
print (m5[2][3])

Output

[5, 6, 7, 8, 9]
13

You may declare a 3-dimensional array by using the data type tensor3 in place of tensor5, a 4-dimensional array using the data type tensor4, and so on up to tensor7.

Plural Constructors

Sometimes, you may want to create variables of the same type in a single declaration. You can do so by using the following syntax −

Syntax

from theano.tensor import * x, y, z = dmatrices('x', 'y', 'z') 
x = ([1,2],[3,4],[5,6]) 
y = ([7,8],[9,10],[11,12]) 
z = ([13,14],[15,16],[17,18]) 
print (x[2]) 
print (y[1]) 
print (z[0])

Output

[5, 6] 
[9, 10] 
[13, 14]
Advertisements