我有一個基于以下代碼的現有項目：https ://gist.github.com/karpathy/d4dee566867f8291f086 。作為學習 TensorFlow 的一個練習，我試圖將此代碼轉換為使用 TF，但是，許多 API 文檔似乎都期望我要移植的代碼中不存在各種術語的先驗知識。我假設我會使用 SimpleRNN 或 RNN 的某個子類。這段代碼的基本 TensorFlow 版本會是什么樣子？
1 回答
呼如林
TA貢獻1798條經驗 獲得超3個贊
在獲得了一些空閑時間和大量的擺弄之后，我想出了類似的東西：
import numpy as np
import tensorflow as tf
from tensorflow.keras import layers, initializers
from tensorflow.keras.optimizers import Adagrad
SCALE = 100.0
def makeXVector(ix):
x = np.zeros((vocab_size, ))
x[ix] = 1.0
return x
with open('input.txt', 'r') as f:
print('Reading and tokenising input.txt')
data = f.read()
print('Collecting chars')
chars = list(set(data))
data_size, vocab_size = len(data), len(chars)
char_to_ix = { tok:i for i,tok in enumerate(chars) }
ix_to_char = { i:tok for i,tok in enumerate(chars) }
# hyperparameters
hidden_size = 300 # size of hidden layer of neurons
seq_length = 25 # number of steps to unroll the RNN for
learning_rate = 1e-1
model = tf.keras.Sequential()
model.add(
layers.SimpleRNN(
hidden_size,
activation='tanh',
use_bias=True,
stateful=True,
bias_initializer=initializers.Zeros(),
# (batch_size, timesteps, input_dim)
batch_input_shape=np.array((1, 1, vocab_size))
)
)
model.add(
layers.Dense(
vocab_size,
activation='softmax',
use_bias=True,
bias_initializer=initializers.Zeros(),
)
)
model.compile(
optimizer=Adagrad(learning_rate=learning_rate),
loss='categorical_crossentropy'
)
n, p = 0, 0
smooth_loss = -np.log(1.0/vocab_size)*seq_length # loss at iteration 0
def sample(model, seed_ix, n):
x = makeXVector(seed_ix)
ixes = []
for t in range(n):
p = np.array(model.predict(np.array([[x.ravel()]]))).ravel()
# Select a random character based on the probability
ix = np.random.choice(list(range(vocab_size)), p=p)
x = makeXVector(ix)
ixes.append(ix)
return ixes
print('data has {:d} characters, {:d} unique.'.format(data_size, vocab_size))
try:
while True:
# prepare inputs (we're sweeping from left to right in steps seq_length long)
if p+seq_length+1 >= len(data) or n == 0:
p = 0 # go from start of data
inputs = [char_to_ix[tok] for tok in data[p:p+seq_length]]
targets = [char_to_ix[tok] for tok in data[p+1:p+seq_length+1]]
# sample from the model now and then
if n % 100 == 0:
sample_ix = sample(model, inputs[0], 200)
txt = ' '.join((ix_to_char[ix][0] for ix in sample_ix))
print('----\n{}\n----'.format(txt))
hist = model.fit(
np.array([[inp] for inp in map(makeXVector, inputs)]),
np.array([targ for targ in map(makeXVector, targets)]),
verbose=0
)
loss = hist.history['loss'][-1]
smooth_loss = smooth_loss * 0.999 + loss * 0.001
if n % 100 == 0: print('iter {:d}, loss: {:f}'.format(n, smooth_loss)) # print progress
p += seq_length # move data pointer
n += 1 # iteration counter
except KeyboardInterrupt:
pass
添加回答
舉報
0/150
提交
取消