tensorflow/python/ops/numpy_ops/g3doc/TensorFlow_NumPy_Text_Generation.ipynb
#@title Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
This tutorial demonstrates how to generate text using a character-based RNN. We will work with a dataset of Shakespeare's writing from Andrej Karpathy's The Unreasonable Effectiveness of Recurrent Neural Networks. Given a sequence of characters from this data ("Shakespear"), train a model to predict the next character in the sequence ("e"). Longer sequences of text can be generated by calling the model repeatedly.
Note: Enable GPU acceleration to execute this notebook faster. In Colab: Runtime > Change runtime type > Hardware accelerator > GPU. If running locally make sure TensorFlow version >= 2.4.
This tutorial includes runnable code implemented using tf.experimental.numpy. The following is sample output when the model in this tutorial trained for 30 epochs, and started with the string "Q":
<pre> QUEENE: I had thought thou hadst a Roman; for the oracle, Thus by All bids the man against the word, Which are so weak of care, by old care done; Your children were in your holy love, And the precipitation through the bleeding throne. BISHOP OF ELY: Marry, and will, my lord, to weep in such a one were prettiest; Yet now I was adopted heir Of the world's lamentable day, To watch the next way with his father with his face? ESCALUS: The cause why then we are all resolved more sons. VOLUMNIA: O, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, it is no sin it should be dead, And love and pale as any will to that word. QUEEN ELIZABETH: But how long have I heard the soul for this world, And show his hands of life be proved to stand. PETRUCHIO: I say he look'd on, if I must be content To stay him from the fatal of our country's bliss. His lordship pluck'd from this sentence then for prey, And then let us twain, being the moon, were she such a case as fills m </pre>While some of the sentences are grammatical, most do not make sense. The model has not learned the meaning of words, but consider:
The model is character-based. When training started, the model did not know how to spell an English word, or that words were even a unit of text.
The structure of the output resembles a play—blocks of text generally begin with a speaker name, in all capital letters similar to the dataset.
As demonstrated below, the model is trained on small batches of text (100 characters each), and is still able to generate a longer sequence of text with coherent structure.
import tensorflow as tf
import tensorflow.experimental.numpy as tnp
import numpy as np
import os
import time
Change the following line to run this code on your own data.
path_to_file = tf.keras.utils.get_file('shakespeare.txt', 'https://storage.googleapis.com/download.tensorflow.org/data/shakespeare.txt')
First, look in the text:
# Read, then decode for py2 compat.
text = open(path_to_file, 'rb').read().decode(encoding='utf-8')
# length of text is the number of characters in it
print ('Length of text: {} characters'.format(len(text)))
# Take a look at the first 250 characters in text
print(text[:250])
# The unique characters in the file
vocab = sorted(set(text))
print ('{} unique characters'.format(len(vocab)))
Before training, we need to map strings to a numerical representation. Create two lookup tables: one mapping characters to numbers, and another for numbers to characters.
# Creating a mapping from unique characters to indices
char2idx = {u:i for i, u in enumerate(vocab)}
idx2char = np.array(vocab)
text_as_int = np.array([char2idx[c] for c in text])
Given a character, or a sequence of characters, what is the most probable next character? This is the task we're training the model to perform. The input to the model will be a sequence of characters, and we train the model to predict the output—the following character at each time step.
Since RNNs maintain an internal state that depends on the previously seen elements, given all the characters computed until this moment, what is the next character?
Next divide the text into example sequences. Each input sequence will contain seq_length characters from the text.
For each input sequence, the corresponding targets contain the same length of text, except shifted one character to the right.
So break the text into chunks of seq_length+1. For example, say seq_length is 4 and our text is "Hello". The input sequence would be "Hell", and the target sequence "ello".
# The maximum length sentence we want for a single input in characters
seq_length = 100
examples_per_epoch = len(text)//(seq_length+1)
# Create training examples / targets
char_dataset = tf.data.Dataset.from_tensor_slices(text_as_int)
for i in char_dataset.take(5):
print(idx2char[i.numpy()])
sequences = char_dataset.batch(seq_length+1, drop_remainder=True)
for item in sequences.take(5):
print(repr(''.join(idx2char[item.numpy()])))
def split_input_target(chunk):
input_text = chunk[:-1]
target_text = chunk[1:]
return input_text, target_text
dataset = sequences.map(split_input_target)
for input_example, target_example in dataset.take(1):
print ('Input data: ', repr(''.join(idx2char[input_example.numpy()])))
print ('Target data:', repr(''.join(idx2char[target_example.numpy()])))
Each index of these vectors are processed as one time step. For the input at time step 0, the model receives the index for "F" and tries to predict the index for "i" as the next character. At the next timestep, it does the same thing but the RNN considers the previous step context in addition to the current input character.
for i, (input_idx, target_idx) in enumerate(zip(input_example[:5], target_example[:5])):
print("Step {:4d}".format(i))
print(" input: {} ({:s})".format(input_idx, repr(idx2char[input_idx])))
print(" expected output: {} ({:s})".format(target_idx, repr(idx2char[target_idx])))
We used tf.data to split the text into manageable sequences. But before feeding this data into the model, we need to shuffle the data and pack it into batches.
# Batch size
BATCH_SIZE = 64
# Buffer size to shuffle the dataset
# (TF data is designed to work with possibly infinite sequences,
# so it doesn't attempt to shuffle the entire sequence in memory. Instead,
# it maintains a buffer in which it shuffles elements).
BUFFER_SIZE = 10000
dataset = dataset.shuffle(BUFFER_SIZE).batch(BATCH_SIZE, drop_remainder=True)
dataset
We manually implement the model from scratch, using tf.numpy and some low-level TF ops. A Model object has three layers: Embedding, GRU and Dense. Embedding and Dense are little more than just wrappers around tnp.take and tnp.dot, but we can use them to familiarize ourself with the structure of a layer. Each layer has two essential methods: build and __call__. build creates and initializes the layer's weights and state, which are things that change during the training process. __call__ is the forward function that calculates outputs given inputs, using the layer's weights and state internally.
Our model (more precisely the GRU layer) is stateful, because each call of __call__ will change its internal state, affecting the next call.
# Length of the vocabulary in chars
vocab_size = len(vocab)
# The embedding dimension
embedding_dim = 256
# Number of RNN units
rnn_units = 1024
class Embedding:
def __init__(self, vocab_size, embedding_dim):
self._vocab_size = vocab_size
self._embedding_dim = embedding_dim
self._built = False
def __call__(self, inputs):
if not self._built:
self.build(inputs)
return tnp.take(self.weights, inputs, axis=0)
def build(self, inputs):
del inputs
self.weights = tf.Variable(tnp.random.randn(
self._vocab_size, self._embedding_dim).astype(np.float32))
self._built = True
class GRUCell:
"""Builds a traditional GRU cell with dense internal transformations.
Gated Recurrent Unit paper: https://arxiv.org/abs/1412.3555
"""
def __init__(self, n_units, forget_bias=0.0):
self._n_units = n_units
self._forget_bias = forget_bias
self._built = False
def __call__(self, inputs):
if not self._built:
self.build(inputs)
x, gru_state = inputs
# Dense layer on the concatenation of x and h.
y = tnp.dot(tnp.concatenate([x, gru_state], axis=-1), self.w1) + self.b1
# Update and reset gates.
u, r = tnp.split(tf.sigmoid(y), 2, axis=-1)
# Candidate.
c = tnp.dot(tnp.concatenate([x, r * gru_state], axis=-1), self.w2) + self.b2
new_gru_state = u * gru_state + (1 - u) * tnp.tanh(c)
return new_gru_state
def build(self, inputs):
# State last dimension must be n_units.
assert inputs[1].shape[-1] == self._n_units
# The dense layer input is the input and half of the GRU state.
dense_shape = inputs[0].shape[-1] + self._n_units
self.w1 = tf.Variable(tnp.random.uniform(
-0.01, 0.01, (dense_shape, 2 * self._n_units)).astype(tnp.float32))
self.b1 = tf.Variable((tnp.random.randn(2 * self._n_units) * 1e-6 + self._forget_bias
).astype(tnp.float32))
self.w2 = tf.Variable(tnp.random.uniform(
-0.01, 0.01, (dense_shape, self._n_units)).astype(tnp.float32))
self.b2 = tf.Variable((tnp.random.randn(self._n_units) * 1e-6).astype(tnp.float32))
self._built = True
@property
def weights(self):
return (self.w1, self.b1, self.w2, self.b2)
class GRU:
def __init__(self, n_units, forget_bias=0.0, stateful=False):
self._cell = GRUCell(n_units, forget_bias)
self._stateful = stateful
self._built = False
def __call__(self, inputs):
if not self._built:
self.build(inputs)
if self._stateful:
state = self.state.read_value()
else:
state = self._init_state(inputs.shape[0])
inputs = tnp.transpose(inputs, (1, 0, 2))
output = tf.scan(
lambda gru_state, x: self._cell((x, gru_state)),
inputs, state)
if self._stateful:
self.state.assign(output[-1, ...])
return tnp.transpose(output, [1, 0, 2])
def _init_state(self, batch_size):
return tnp.zeros([batch_size, self._cell._n_units], tnp.float32)
def reset_state(self):
if not self._stateful:
return
self.state.assign(tf.zeros_like(self.state))
def create_state(self, batch_size):
self.state = tf.Variable(self._init_state(batch_size))
def build(self, inputs):
s = inputs.shape[0:1] + inputs.shape[2:]
shapes = (s, s[:-1] + (self._cell._n_units,))
self._cell.build([tf.TensorSpec(x, tf.float32) for x in shapes])
if self._stateful:
self.create_state(inputs.shape[0])
else:
self.state = ()
self._built = True
@property
def weights(self):
return self._cell.weights
class Dense:
def __init__(self, n_units, activation=None):
self._n_units = n_units
self._activation = activation
self._built = False
def __call__(self, inputs):
if not self._built:
self.build(inputs)
y = tnp.dot(inputs, self.w) +self.b
if self._activation != None:
y = self._activation(y)
return y
def build(self, inputs):
shape_w = (inputs.shape[-1], self._n_units)
lim = tnp.sqrt(6.0 / (shape_w[0] + shape_w[1]))
self.w = tf.Variable(tnp.random.uniform(-lim, lim, shape_w).astype(tnp.float32))
self.b = tf.Variable((tnp.random.randn(self._n_units) * 1e-6).astype(tnp.float32))
self._built = True
@property
def weights(self):
return (self.w, self.b)
class Model:
def __init__(self, vocab_size, embedding_dim, rnn_units, forget_bias=0.0, stateful=False, activation=None):
self._embedding = Embedding(vocab_size, embedding_dim)
self._gru = GRU(rnn_units, forget_bias=forget_bias, stateful=stateful)
self._dense = Dense(vocab_size, activation=activation)
self._layers = [self._embedding, self._gru, self._dense]
self._built = False
def __call__(self, inputs):
if not self._built:
self.build(inputs)
xs = inputs
for layer in self._layers:
xs = layer(xs)
return xs
def build(self, inputs):
self._embedding.build(inputs)
self._gru.build(tf.TensorSpec(inputs.shape + (self._embedding._embedding_dim,), tf.float32))
self._dense.build(tf.TensorSpec(inputs.shape + (self._gru._cell._n_units,), tf.float32))
self._built = True
@property
def weights(self):
return [layer.weights for layer in self._layers]
@property
def state(self):
return self._gru.state
def create_state(self, *args):
self._gru.create_state(*args)
def reset_state(self, *args):
self._gru.reset_state(*args)
model = Model(
vocab_size = vocab_size,
embedding_dim=embedding_dim,
rnn_units=rnn_units,
stateful=True)
For each character the model looks up the embedding, runs the GRU one timestep with the embedding as input, and applies the dense layer to generate logits predicting the log-likelihood of the next character.
Now run the model to see that it behaves as expected.
First check the shape of the output:
for input_example_batch, target_example_batch in dataset.take(1):
input_example_batch = tnp.asarray(input_example_batch)
example_batch_predictions = model(input_example_batch)
print(example_batch_predictions.shape, "# (batch_size, sequence_length, vocab_size)")
In the above example the sequence length of the input is 100 but the model can be run on inputs of any length:
To get actual predictions from the model we need to sample from the output distribution, to get actual character indices. This distribution is defined by the logits over the character vocabulary.
Note: It is important to sample from this distribution as taking the argmax of the distribution can easily get the model stuck in a loop.
Try it for the first example in the batch:
example_batch_predictions[0]
sampled_indices = tf.random.categorical(example_batch_predictions[0], num_samples=1)
sampled_indices = tf.squeeze(sampled_indices,axis=-1).numpy()
This gives us, at each timestep, a prediction of the next character index:
sampled_indices
Decode these to see the text predicted by this untrained model:
print("Input: \n", repr("".join(idx2char[input_example_batch[0]])))
print()
print("Next Char Predictions: \n", repr("".join(idx2char[sampled_indices ])))
At this point the problem can be treated as a standard classification problem. Given the previous RNN state, and the input this time step, predict the class of the next character.
We define the loss function from scratch, using tf.nn.log_softmax. (Our definition is the same as tf.keras.losses.sparse_categorical_crossentropy.)
def one_hot(labels, n):
return (labels[..., np.newaxis] == tnp.arange(n)).astype(np.float32)
def loss_fn(labels, predictions):
predictions = tf.nn.log_softmax(predictions)
return -tnp.sum(predictions * one_hot(tnp.asarray(labels), predictions.shape[-1]), axis=-1)
example_batch_loss = loss_fn(target_example_batch, example_batch_predictions)
print("Prediction shape: ", example_batch_predictions.shape, " # (batch_size, sequence_length, vocab_size)")
print("scalar_loss: ", tnp.mean(example_batch_loss))
Keeping the DIY spirit, we implement the Adam optimizer from scratch.
class Adam:
def __init__(self, learning_rate=0.001, b1=0.9, b2=0.999, eps=1e-7):
self._lr = learning_rate
self._b1 = b1
self._b2 = b2
self._eps = eps
self._built = False
def build(self, weights):
self._m = tf.nest.map_structure(lambda x: tf.Variable(tnp.zeros_like(x)), weights)
self._v = tf.nest.map_structure(lambda x: tf.Variable(tnp.zeros_like(x)), weights)
self._step = tf.Variable(tnp.asarray(0, np.int64))
self._built = True
def _update(self, weights_var, grads, m_var, v_var):
b1 = self._b1
b2 = self._b2
eps = self._eps
step = tnp.asarray(self._step, np.float32)
lr = self._lr
weights = tnp.asarray(weights_var)
m = tnp.asarray(m_var)
v = tnp.asarray(v_var)
m = (1 - b1) * grads + b1 * m # First moment estimate.
v = (1 - b2) * (grads ** 2) + b2 * v # Second moment estimate.
mhat = m / (1 - b1 ** (step + 1)) # Bias correction.
vhat = v / (1 - b2 ** (step + 1))
weights_var.assign_sub((lr * mhat / (tnp.sqrt(vhat) + eps)).astype(weights.dtype))
m_var.assign(m)
v_var.assign(v)
def apply_gradients(self, weights, grads):
if not self._built:
self.build(weights)
tf.nest.map_structure(lambda *args: self._update(*args), weights, grads, self._m, self._v)
self._step.assign_add(1)
@property
def state(self):
return (self._step, self._m, self._v)
optimizer = Adam()
Again, we write our training loop from scratch.
To keep training time reasonable, use 10 epochs to train the model. In Colab, set the runtime to GPU for faster training.
@tf.function
def train_step(inp, target):
with tf.GradientTape() as tape:
# tape.watch(tf.nest.flatten(weights))
predictions = model(inp)
loss = tnp.mean(loss_fn(target, predictions))
weights = model.weights
grads = tape.gradient(loss, weights)
optimizer.apply_gradients(weights, grads)
return loss
# Training step
EPOCHS = 10
model.create_state(BATCH_SIZE)
for epoch in range(EPOCHS):
start = time.time()
# initializing the hidden state at the start of every epoch
model.reset_state()
for (batch_n, (inp, target)) in enumerate(dataset):
loss = train_step(inp, target)
if batch_n % 100 == 0:
template = 'Epoch {} Batch {} Loss {}'
print(template.format(epoch+1, batch_n, loss))
print ('Epoch {} Loss {}'.format(epoch+1, loss))
print ('Time taken for 1 epoch {} sec\n'.format(time.time() - start))
The following code block generates the text:
It Starts by choosing a start string, initializing the RNN state and setting the number of characters to generate.
Get the prediction distribution of the next character using the start string and the RNN state.
Then, use a categorical distribution to calculate the index of the predicted character. Use this predicted character as our next input to the model.
The RNN state returned by the model is fed back into the model so that it now has more context, instead than only one character. After predicting the next character, the modified RNN states are again fed back into the model, which is how it learns as it gets more context from the previously predicted characters.
Looking at the generated text, you'll see the model knows when to capitalize, make paragraphs and imitates a Shakespeare-like writing vocabulary. With the small number of training epochs, it has not yet learned to form coherent sentences.
To keep this prediction step simple, use a batch size of 1.
def generate_text(model, start_string):
# Evaluation step (generating text using the learned model)
# Number of characters to generate
num_generate = 1000
# Converting our start string to numbers (vectorizing)
input_eval = [char2idx[s] for s in start_string]
input_eval = tf.expand_dims(input_eval, 0)
# Empty string to store our results
text_generated = []
# Low temperatures results in more predictable text.
# Higher temperatures results in more surprising text.
# Experiment to find the best setting.
temperature = 1.0
# Here batch size == 1
model.create_state(1)
for i in range(num_generate):
predictions = model(input_eval)
# remove the batch dimension
predictions = tf.squeeze(predictions, 0)
# using a categorical distribution to predict the character returned by the model
predictions = predictions / temperature
predicted_id = tf.random.categorical(predictions, num_samples=1)[-1,0].numpy()
# We pass the predicted character as the next input to the model
# along with the previous hidden state
input_eval = tf.expand_dims([predicted_id], 0)
text_generated.append(idx2char[predicted_id])
return (start_string + ''.join(text_generated))
print(generate_text(model, start_string=u"ROMEO: "))
The easiest thing you can do to improve the results it to train it for longer (try EPOCHS=30).
You can also experiment with a different start string, or try adding another RNN layer to improve the model's accuracy, or adjusting the temperature parameter to generate more or less random predictions.