network with Py Torch Prediction CharRNN Prediction Flow

network with Py. Torch (Prediction)

Char_RNN • Prediction Flow • sample(net, 1000, ‘Anna’, 5) (model, size, prime, top_k) sample GPU Check predict for: prime predict for: size 2

Char_RNN • Prediction Flow • predict(net, ‘A’, h, 5) (model, char, h, top_k) predict Char to One-hot vector Model Softmax topk random. choice 3

Char_RNN • def sample • sample(net, 1000, ‘Anna’, 5) 4

Char_RNN • def sample • sample(net, 1000, ‘Anna’, 5) chars A predict n n a , while: prime 5

Char_RNN • def sample • sample(net, 1000, ‘Anna’, 5) chars A n n a , ‘ ‘. join predict while: size 6

Char_RNN • def predict • predict(net, ‘A’, h, 5) 7

Char_RNN • def predict x : char 의 onehotvector inputs : x의 type이 tensor Char to Onehotvector 8

Char_RNN • def predict 9

Char_RNN • def predict 10

Char_RNN • def predict Model 11

Char_RNN • def predict 43 21 33 81 17 22 … 45 82 28 (1, 83) Softmax 12

Char_RNN • def predict Softmax 13

Char_RNN • sum with dim T 1 T 2 T 3 T 4 Sum(dim=0) (2, 2) T 1 (1, 2) T 2 T 3 T 4 (2, 2) T 1+T 3 T 2+T 4 Sum(dim=1) T 1+T 2 T 3+T 4 (2, 1) 14/27

Char_RNN • Softmax dim 15/27

Char_RNN • def predict • top_k = 5 topk 16

Char_RNN • topk • Example top_k = 1 17

Char_RNN • def predict • top_k = 5 randomchoice 18

Char_RNN • def predict • top_k = 5 random. choice randomchoice top_ch 14 33 76 25 70 p 0. 3629 0. 1676 0. 1318 0. 0714 0. 0398 19

Char_RNN • 1005 chars A n n 5 a , w i t 1, 000 h … r 20

Char_RNN • 1005 chars 21

Char_RNN • 1005 chars • ’ ’. join() 22

Thank You

Char_RNN • topk • top_k = 5 24

Char_RNN • sum with dim • 3 d 인 경우 (2, 2, 2) T 1 T 3 T 5 T 2 T 7 T 4 T 6 T 8 T 8 Sum(dim=0) T 1+T 5 T 2+T 6 T 3+T 7 T 4+T 8 Sum(dim=1) T 1+T 3 T 2+T 4 T 5+T 7 T 6+T 8 Sum(dim=2) (2, 2) T 1+T 2 T 3+T 4 T 5+T 6 T 7+T 8 (2, 2)

Char_RNN • Softmax with dim default • 2 d 인 경우 • Default dimension : 1 • Softmax( T 1 T 2 T 3 T 4 ) = T 1/(T 1+T 2) T 2/(T 1+T 2) T 3/(T 3+T 4) T 4/(T 3+T 4) (2, 2) • 3 d 인 경우 • Default dimension : 0 T 1 • Softmax( T 3 T 2 T 5 T 4 T 7 (2, 2, 2) T 6 T 8 ) = T 1/(T 1+T 5) T 2/(T 2+T 6) T 5/(T 1+T 5) T 6/(T 2+T 6) T 3/(T 3+T 7) T 4/(T 4+T 8) T 7/(T 3+T 7) T 8/(T 4+T 8) (2, 2, 2) 26/27

Char_RNN • Softmax with dim • 2 d 인 경우 • dim : 0 • Softmax( T 1 T 2 T 3 T 4 (2, 2) ) = T 1/(T 1+T 3 T 2/(T 2+T 4 ) ) T 3/(T 1+T 3 T 4/(T 2+T 4 ) (2, 2) 27/27

Char_RNN • Softmax with dim • 3 d 인 경우 • dim : 1 T 1 • Softmax( T 3 T 2 T 5 T 4 T 7 T 6 ) = T 8 (2, 2, 2) T 1/(T 1+T 3) T 2/(T 2+T 4) T 5/(T 5+T 7) T 6/(T 6+T 8) T 3/(T 1+T 3) T 4/(T 2+T 4) T 7/(T 5+T 7) T 8/(T 6+T 8) (2, 2, 2) • dim : 2 T 1 • Softmax( T 3 T 2 T 5 T 4 T 7 (2, 2, 2) T 6 T 8 ) = T 1/(T 1+T 2) T 2/(T 1+T 2) T 5/(T 5+T 6) T 6/(T 5+T 6) T 3/(T 3+T 4) T 4/(T 3+T 4) T 7/(T 7+T 8) T 8/(T 7+T 8) (2, 2, 2) 28/27