텐서플로와 파이썬 자료형
DT_FLOAT |
tf.float32 |
32비트 실수 |
DT_DOUBLE |
tf.float64 |
64비트 실수 |
DT_INT8 |
tf.int8 |
8비트 정수 |
DT_INT16 |
tf.int16 |
16비트 정수 |
DT_INT32 |
tf.int32 |
32비트 정수 |
DT_INT64 |
tf.int64 |
64비트 정수 |
DT_UINT8 |
tf.uint8 |
8비트 부호없는 정수 |
DT_UINT16 |
tf.uint16 |
16비트 부호없는 정수 |
DT_DTSTRING |
tf.string |
문자형 |
DT_BOOL |
tf.bool |
논리형 |
DT_COMPLEX664 |
tf.complex64 |
64비트 복소수. 실수파트 32비트, 복소수파트 32비트 |
DT_COMPLEX128 |
tf.complex128 |
128비트 복소수. 실수파트 64비트, 복소수파트 64비트 |
DT_QINT8 |
tf.qint8 |
양자화 계산에 사용되는 8비트 정수 |
DT_QINT32 |
tf.qint32 |
양자화 계산에 사용되는 32비트 정수 |
DT_QUINT8 |
tf.quint8 |
양자화 계산에 사용되는 부호없는 8비트 정수 |
랭크, 파이썬 예제, R 예제
0 |
Scalar (magnitude only) |
s=483 |
s<-483 |
1 |
Vector (magnitude and direction) |
v=[1.1,2.2,3.3] |
v<-c(1.1,2.2,3.3) |
2 |
Matrix (table of numbers) |
m=[[1,2,3],[4,5,6],[7,8,9]] |
m<-1:9 ; dim(m)<-c(3,3) |
3 |
3-Tensor (cube of numbers) |
t=[[[[2],[4],[6]],[[8],[10],[12]],[[14],[16],[18]]] |
t<-2*(1:9) ; dim(t)<-c(1,3,3) |
n |
n-Tensor (you get the idea) |
… |
… |
구조, 랭크, 차원번호, 예제
0 |
[] |
0-D |
A 0-D tensor. A scalar. |
1 |
[D0] |
1-D |
A 1-D tensor with shape[5]. |
2 |
[D0,D1] |
2-D |
A 2-D tensor with shape [3,4]. |
3 |
[D0,D1,D2] |
3-D |
A 3-D tensor with shape [1,4,3]. |
n |
[D0,D1,…,Dn-1] |
n-D |
A tensor with shape [D0,D1,…,Dn-1]. |
API Reference
Scalar |
Single-element vector |
1, 1L, TRUE, “foo” |
List |
Multi-element vector |
c(1.0, 2.0, 3.0), c(1L, 2L, 3L) |
Tuple |
List of multiple types |
list(1L, TRUE, “foo”) |
Dict |
Named list or dict |
list(a=1L, b=2.0), dict(x=x_data) |
NumPy ndarray |
Matrix/Array |
matrix(c(1,2,3,4), nrow=2, ncol=2) |
None, True, False |
NULL, TRUE, FALSE |
NULL, TRUE, FALSE |