In this project, you'll generate your own Seinfeld TV scripts using RNNs. You'll be using part of the Seinfeld dataset of scripts from 9 seasons. The Neural Network you'll build will generate a new ,"fake" TV script, based on patterns it recognizes in this training data.
The data is already provided for you in ./data/Seinfeld_Scripts.txt and you're encouraged to open that file and look at the text.
- As a first step, we'll load in this data and look at some samples.
- Then, you'll be tasked with defining and training an RNN to generate a new script!
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# load in data
import helper
data_dir = './data/Seinfeld_Scripts.txt'
text = helper.load_data(data_dir)Play around with view_line_range to view different parts of the data. This will give you a sense of the data you'll be working with. You can see, for example, that it is all lowercase text, and each new line of dialogue is separated by a newline character \n.
view_line_range = (0, 10)
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
import numpy as np
print('Dataset Stats')
print('Roughly the number of unique words: {}'.format(len({word: None for word in text.split()})))
lines = text.split('\n')
print('Number of lines: {}'.format(len(lines)))
word_count_line = [len(line.split()) for line in lines]
print('Average number of words in each line: {}'.format(np.average(word_count_line)))
print()
print('The lines {} to {}:'.format(*view_line_range))
print('\n'.join(text.split('\n')[view_line_range[0]:view_line_range[1]]))Dataset Stats
Roughly the number of unique words: 46367
Number of lines: 109233
Average number of words in each line: 5.544240293684143
The lines 0 to 10:
jerry: do you know what this is all about? do you know, why were here? to be out, this is out...and out is one of the single most enjoyable experiences of life. people...did you ever hear people talking about we should go out? this is what theyre talking about...this whole thing, were all out now, no one is home. not one person here is home, were all out! there are people trying to find us, they dont know where we are. (on an imaginary phone) did you ring?, i cant find him. where did he go? he didnt tell me where he was going. he must have gone out. you wanna go out you get ready, you pick out the clothes, right? you take the shower, you get all ready, get the cash, get your friends, the car, the spot, the reservation...then youre standing around, what do you do? you go we gotta be getting back. once youre out, you wanna get back! you wanna go to sleep, you wanna get up, you wanna go out again tomorrow, right? where ever you are in life, its my feeling, youve gotta go.
jerry: (pointing at georges shirt) see, to me, that button is in the worst possible spot. the second button literally makes or breaks the shirt, look at it. its too high! its in no-mans-land. you look like you live with your mother.
george: are you through?
jerry: you do of course try on, when you buy?
george: yes, it was purple, i liked it, i dont actually recall considering the buttons.
The first thing to do to any dataset is pre-processing. Implement the following pre-processing functions below:
- Lookup Table
- Tokenize Punctuation
To create a word embedding, you first need to transform the words to ids. In this function, create two dictionaries:
- Dictionary to go from the words to an id, we'll call
vocab_to_int - Dictionary to go from the id to word, we'll call
int_to_vocab
Return these dictionaries in the following tuple (vocab_to_int, int_to_vocab)
import problem_unittests as tests
from collections import Counter
def create_lookup_tables(text):
"""
Create lookup tables for vocabulary
:param text: The text of tv scripts split into words
:return: A tuple of dicts (vocab_to_int, int_to_vocab)
"""
# TODO: Implement Function
wCounts = Counter(text)
sVocab = sorted(wCounts, key=wCounts.get, reverse=True)
int_to_vocab = {ii: word for ii, word in enumerate(sVocab)}
vocab_to_int = {word: ii for ii, word in int_to_vocab.items()}
# return tuple
return (vocab_to_int, int_to_vocab)
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_create_lookup_tables(create_lookup_tables)Tests Passed
We'll be splitting the script into a word array using spaces as delimiters. However, punctuations like periods and exclamation marks can create multiple ids for the same word. For example, "bye" and "bye!" would generate two different word ids.
Implement the function token_lookup to return a dict that will be used to tokenize symbols like "!" into "||Exclamation_Mark||". Create a dictionary for the following symbols where the symbol is the key and value is the token:
- Period ( . )
- Comma ( , )
- Quotation Mark ( " )
- Semicolon ( ; )
- Exclamation mark ( ! )
- Question mark ( ? )
- Left Parentheses ( ( )
- Right Parentheses ( ) )
- Dash ( - )
- Return ( \n )
This dictionary will be used to tokenize the symbols and add the delimiter (space) around it. This separates each symbols as its own word, making it easier for the neural network to predict the next word. Make sure you don't use a value that could be confused as a word; for example, instead of using the value "dash", try using something like "||dash||".
def token_lookup():
"""
Generate a dict to turn punctuation into a token.
:return: Tokenized dictionary where the key is the punctuation and the value is the token
"""
# TODO: Implement Function
tokens = {
'.':'||PERIOD||',
',':'||COMMA||',
'"':'||QUOTATION_MARK||',
';':'||SEMICOLON||',
'!':'||EXCLAMATION_MARK||',
'?':'||QUESTION_MARK||',
'(':'||LEFT_PAREN||',
')':'||RIGHT_PAREN||',
'-':'||DASH||',
'\n':'||RETURN||'
}
return tokens
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_tokenize(token_lookup)Tests Passed
Running the code cell below will pre-process all the data and save it to file. You're encouraged to lok at the code for preprocess_and_save_data in the helpers.py file to see what it's doing in detail, but you do not need to change this code.
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# pre-process training data
helper.preprocess_and_save_data(data_dir, token_lookup, create_lookup_tables)This is your first checkpoint. If you ever decide to come back to this notebook or have to restart the notebook, you can start from here. The preprocessed data has been saved to disk.
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
import problem_unittests as tests
int_text, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()In this section, you'll build the components necessary to build an RNN by implementing the RNN Module and forward and backpropagation functions.
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import torch
# Check for a GPU
train_on_gpu = torch.cuda.is_available()
if not train_on_gpu:
print('No GPU found. Please use a GPU to train your neural network.')
print(train_on_gpu)True
Let's start with the preprocessed input data. We'll use TensorDataset to provide a known format to our dataset; in combination with DataLoader, it will handle batching, shuffling, and other dataset iteration functions.
You can create data with TensorDataset by passing in feature and target tensors. Then create a DataLoader as usual.
data = TensorDataset(feature_tensors, target_tensors)
data_loader = torch.utils.data.DataLoader(data,
batch_size=batch_size)
Implement the batch_data function to batch words data into chunks of size batch_size using the TensorDataset and DataLoader classes.
You can batch words using the DataLoader, but it will be up to you to create
feature_tensorsandtarget_tensorsof the correct size and content for a givensequence_length.
For example, say we have these as input:
words = [1, 2, 3, 4, 5, 6, 7]
sequence_length = 4
Your first feature_tensor should contain the values:
[1, 2, 3, 4]
And the corresponding target_tensor should just be the next "word"/tokenized word value:
5
This should continue with the second feature_tensor, target_tensor being:
[2, 3, 4, 5] # features
6 # target
from torch.utils.data import TensorDataset, DataLoader
import numpy as np
def batch_data(words, sequence_length, batch_size):
"""
Batch the neural network data using DataLoader
:param words: The word ids of the TV scripts
:param sequence_length: The sequence length of each batch
:param batch_size: The size of each batch; the number of sequences in a batch
:return: DataLoader with batched data
"""
# TODO: Implement function
feature = []
target = []
for i in range(len(words)):
if(len(words[i:i + sequence_length]) == sequence_length):
feature.append(words[i : i + sequence_length])
target.append(words[len(words)-1]) if ((i + sequence_length) >= len(words)) else target.append(words[i + sequence_length])
feature_tensors = np.array(feature)
target_tensors = np.array(target)
data = TensorDataset(torch.from_numpy(feature_tensors), torch.from_numpy(target_tensors))
data_loader = torch.utils.data.DataLoader(data,shuffle=True, batch_size=batch_size)
return data_loader
# there is no test for this function, but you are encouraged to create
# print statements and tests of your ownYou'll have to modify this code to test a batching function, but it should look fairly similar.
Below, we're generating some test text data and defining a dataloader using the function you defined, above. Then, we are getting some sample batch of inputs sample_x and targets sample_y from our dataloader.
Your code should return something like the following (likely in a different order, if you shuffled your data):
torch.Size([10, 5])
tensor([[ 28, 29, 30, 31, 32],
[ 21, 22, 23, 24, 25],
[ 17, 18, 19, 20, 21],
[ 34, 35, 36, 37, 38],
[ 11, 12, 13, 14, 15],
[ 23, 24, 25, 26, 27],
[ 6, 7, 8, 9, 10],
[ 38, 39, 40, 41, 42],
[ 25, 26, 27, 28, 29],
[ 7, 8, 9, 10, 11]])
torch.Size([10])
tensor([ 33, 26, 22, 39, 16, 28, 11, 43, 30, 12])
Your sample_x should be of size (batch_size, sequence_length) or (10, 5) in this case and sample_y should just have one dimension: batch_size (10).
You should also notice that the targets, sample_y, are the next value in the ordered test_text data. So, for an input sequence [ 28, 29, 30, 31, 32] that ends with the value 32, the corresponding output should be 33.
# test dataloader
test_text = range(50)
t_loader = batch_data(test_text, sequence_length=5, batch_size=10)
data_iter = iter(t_loader)
sample_x, sample_y = data_iter.next()
print(sample_x.shape)
print(sample_x)
print()
print(sample_y.shape)
print(sample_y)torch.Size([10, 5])
tensor([[ 42, 43, 44, 45, 46],
[ 12, 13, 14, 15, 16],
[ 21, 22, 23, 24, 25],
[ 26, 27, 28, 29, 30],
[ 9, 10, 11, 12, 13],
[ 5, 6, 7, 8, 9],
[ 27, 28, 29, 30, 31],
[ 22, 23, 24, 25, 26],
[ 44, 45, 46, 47, 48],
[ 41, 42, 43, 44, 45]])
torch.Size([10])
tensor([ 47, 17, 26, 31, 14, 10, 32, 27, 49, 46])
Implement an RNN using PyTorch's Module class. You may choose to use a GRU or an LSTM. To complete the RNN, you'll have to implement the following functions for the class:
__init__- The initialize function.init_hidden- The initialization function for an LSTM/GRU hidden stateforward- Forward propagation function.
The initialize function should create the layers of the neural network and save them to the class. The forward propagation function will use these layers to run forward propagation and generate an output and a hidden state.
The output of this model should be the last batch of word scores after a complete sequence has been processed. That is, for each input sequence of words, we only want to output the word scores for a single, most likely, next word.
- Make sure to stack the outputs of the lstm to pass to your fully-connected layer, you can do this with
lstm_output = lstm_output.contiguous().view(-1, self.hidden_dim) - You can get the last batch of word scores by shaping the output of the final, fully-connected layer like so:
# reshape into (batch_size, seq_length, output_size)
output = output.view(batch_size, -1, self.output_size)
# get last batch
out = output[:, -1]
import torch.nn as nn
class RNN(nn.Module):
def __init__(self, vocab_size, output_size, embedding_dim, hidden_dim, n_layers, dropout=0.5):
"""
Initialize the PyTorch RNN Module
:param vocab_size: The number of input dimensions of the neural network (the size of the vocabulary)
:param output_size: The number of output dimensions of the neural network
:param embedding_dim: The size of embeddings, should you choose to use them
:param hidden_dim: The size of the hidden layer outputs
:param dropout: dropout to add in between LSTM/GRU layers
"""
super(RNN, self).__init__()
# TODO: Implement function
# set class variables
self.output_size = output_size
self.n_layers = n_layers
self.hidden_dim = hidden_dim
self.embedding = nn.Embedding(vocab_size, embedding_dim)
self.lstm = nn.LSTM(embedding_dim, hidden_dim, n_layers,
dropout=dropout, batch_first=True)
#self.dropout = nn.Dropout(0.3)
# define model layers
self.fc = nn.Linear(hidden_dim, output_size)
#self.sig = nn.Sigmoid()
def forward(self, nn_input, hidden):
"""
Forward propagation of the neural network
:param nn_input: The input to the neural network
:param hidden: The hidden state
:return: Two Tensors, the output of the neural network and the latest hidden state
"""
# TODO: Implement function
batch_size = nn_input.size(0)
embeds = self.embedding(nn_input)
lstm_out, hidden = self.lstm(embeds, hidden)
lstm_out = lstm_out.contiguous().view(-1, self.hidden_dim)
#out = self.dropout(lstm_out)
out = self.fc(lstm_out)
sig_out = out
sig_out = sig_out.view(batch_size, -1,self.output_size)
sig_out = sig_out[:, -1]
# return one batch of output word scores and the hidden state
return sig_out, hidden
def init_hidden(self, batch_size):
'''
Initialize the hidden state of an LSTM/GRU
:param batch_size: The batch_size of the hidden state
:return: hidden state of dims (n_layers, batch_size, hidden_dim)
'''
# Implement function
# initialize hidden state with zero weights, and move to GPU if available
weight = next(self.parameters()).data
if (train_on_gpu):
hidden = (weight.new(self.n_layers, batch_size, self.hidden_dim).zero_().cuda(),
weight.new(self.n_layers, batch_size, self.hidden_dim).zero_().cuda())
else:
hidden = (weight.new(self.n_layers, batch_size, self.hidden_dim).zero_(),
weight.new(self.n_layers, batch_size, self.hidden_dim).zero_())
return hidden
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_rnn(RNN, train_on_gpu)Tests Passed
Use the RNN class you implemented to apply forward and back propagation. This function will be called, iteratively, in the training loop as follows:
loss = forward_back_prop(decoder, decoder_optimizer, criterion, inp, target)
And it should return the average loss over a batch and the hidden state returned by a call to RNN(inp, hidden). Recall that you can get this loss by computing it, as usual, and calling loss.item().
If a GPU is available, you should move your data to that GPU device, here.
def forward_back_prop(rnn, optimizer, criterion, inp, target, hidden):
"""
Forward and backward propagation on the neural network
:param decoder: The PyTorch Module that holds the neural network
:param decoder_optimizer: The PyTorch optimizer for the neural network
:param criterion: The PyTorch loss function
:param inp: A batch of input to the neural network
:param target: The target output for the batch of input
:return: The loss and the latest hidden state Tensor
"""
# TODO: Implement Function
# move data to GPU, if available
if(train_on_gpu):
rnn.cuda()
inp, target = inp.cuda(), target.cuda()
rnn.train()
hidden = tuple([each.data for each in hidden])
# zero accumulated gradients
rnn.zero_grad()
# get the output from the model
output, hidden = rnn(inp, hidden)
# perform backpropagation and optimization
# calculate the loss and perform backprop
loss = criterion(output.squeeze(), target.long())
loss.backward()
# `clip_grad_norm` helps prevent the exploding gradient problem in RNNs / LSTMs.
clip=5 # gradient clipping
nn.utils.clip_grad_norm_(rnn.parameters(), clip)
optimizer.step()
# return the loss over a batch and the hidden state produced by our model
return loss.item(), hidden
# Note that these tests aren't completely extensive.
# they are here to act as general checks on the expected outputs of your functions
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_forward_back_prop(RNN, forward_back_prop, train_on_gpu)Tests Passed
With the structure of the network complete and data ready to be fed in the neural network, it's time to train it.
The training loop is implemented for you in the train_decoder function. This function will train the network over all the batches for the number of epochs given. The model progress will be shown every number of batches. This number is set with the show_every_n_batches parameter. You'll set this parameter along with other parameters in the next section.
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
def train_rnn(rnn, batch_size, optimizer, criterion, n_epochs, show_every_n_batches=100):
batch_losses = []
rnn.train()
print("Training for %d epoch(s)..." % n_epochs)
for epoch_i in range(1, n_epochs + 1):
# initialize hidden state
hidden = rnn.init_hidden(batch_size)
for batch_i, (inputs, labels) in enumerate(train_loader, 1):
# make sure you iterate over completely full batches, only
n_batches = len(train_loader.dataset)//batch_size
if(batch_i > n_batches):
break
# forward, back prop
loss, hidden = forward_back_prop(rnn, optimizer, criterion, inputs, labels, hidden)
# record loss
batch_losses.append(loss)
# printing loss stats
if batch_i % show_every_n_batches == 0:
print('Epoch: {:>4}/{:<4} Loss: {}\n'.format(
epoch_i, n_epochs, np.average(batch_losses)))
batch_losses = []
# returns a trained rnn
return rnnSet and train the neural network with the following parameters:
- Set
sequence_lengthto the length of a sequence. - Set
batch_sizeto the batch size. - Set
num_epochsto the number of epochs to train for. - Set
learning_rateto the learning rate for an Adam optimizer. - Set
vocab_sizeto the number of uniqe tokens in our vocabulary. - Set
output_sizeto the desired size of the output. - Set
embedding_dimto the embedding dimension; smaller than the vocab_size. - Set
hidden_dimto the hidden dimension of your RNN. - Set
n_layersto the number of layers/cells in your RNN. - Set
show_every_n_batchesto the number of batches at which the neural network should print progress.
If the network isn't getting the desired results, tweak these parameters and/or the layers in the RNN class.
# Data params
# Sequence Length
sequence_length =10 # of words in a sequence
# Batch Size
batch_size =256
# data loader - do not change
train_loader = batch_data(int_text, sequence_length, batch_size)# Training parameters
# Number of Epochs
num_epochs = 5
# Learning Rate
learning_rate = 0.001
# Model parameters
# Vocab size
print(len(int_to_vocab))
vocab_size = len(int_to_vocab)
# Output size
output_size = len(int_to_vocab)
# Embedding Dimension
embedding_dim = 463
# Hidden Dimension
hidden_dim = 512
# Number of RNN Layers
n_layers = 2
# Show stats for every n number of batches
show_every_n_batches = 50021388
In the next cell, you'll train the neural network on the pre-processed data. If you have a hard time getting a good loss, you may consider changing your hyperparameters. In general, you may get better results with larger hidden and n_layer dimensions, but larger models take a longer time to train.
You should aim for a loss less than 3.5.
You should also experiment with different sequence lengths, which determine the size of the long range dependencies that a model can learn.
from workspace_utils import active_session
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
# create model and move to gpu if available
rnn = RNN(vocab_size, output_size, embedding_dim, hidden_dim, n_layers, dropout=0.5)
if train_on_gpu:
rnn.cuda()
# defining loss and optimization functions for training
optimizer = torch.optim.Adam(rnn.parameters(), lr=learning_rate)
criterion = nn.CrossEntropyLoss()
# training the model
with active_session():
trained_rnn = train_rnn(rnn, batch_size, optimizer, criterion, num_epochs, show_every_n_batches)
# saving the trained model
helper.save_model('./save/trained_rnn', trained_rnn)
print('Model Trained and Saved')Training for 5 epoch(s)...
Epoch: 1/5 Loss: 5.115768946647644
Epoch: 1/5 Loss: 4.4893813614845275
Epoch: 1/5 Loss: 4.3190711731910705
Epoch: 1/5 Loss: 4.221667974472046
Epoch: 1/5 Loss: 4.124557318687439
Epoch: 1/5 Loss: 4.078655934333801
Epoch: 2/5 Loss: 3.9476230275340196
Epoch: 2/5 Loss: 3.8329451603889466
Epoch: 2/5 Loss: 3.827289001941681
Epoch: 2/5 Loss: 3.8161864466667175
Epoch: 2/5 Loss: 3.7902850456237793
Epoch: 2/5 Loss: 3.79382661819458
Epoch: 3/5 Loss: 3.680831721400827
Epoch: 3/5 Loss: 3.5935067834854126
Epoch: 3/5 Loss: 3.5977722783088684
Epoch: 3/5 Loss: 3.59462979221344
Epoch: 3/5 Loss: 3.599617018222809
Epoch: 3/5 Loss: 3.5838846497535704
Epoch: 4/5 Loss: 3.4988145784633917
Epoch: 4/5 Loss: 3.422701789855957
Epoch: 4/5 Loss: 3.4170361709594728
Epoch: 4/5 Loss: 3.4336008672714233
Epoch: 4/5 Loss: 3.4467914662361143
Epoch: 4/5 Loss: 3.448900179386139
Epoch: 5/5 Loss: 3.3514625580330204
Epoch: 5/5 Loss: 3.2749983978271486
Epoch: 5/5 Loss: 3.289450369358063
Epoch: 5/5 Loss: 3.3070807356834413
Epoch: 5/5 Loss: 3.3176130170822145
Epoch: 5/5 Loss: 3.3260426087379455
/opt/conda/lib/python3.6/site-packages/torch/serialization.py:193: UserWarning: Couldn't retrieve source code for container of type RNN. It won't be checked for correctness upon loading.
"type " + obj.__name__ + ". It won't be checked "
Model Trained and Saved
For example, did you try different sequence_lengths and find that one size made the model converge faster? What about your hidden_dim and n_layers; how did you decide on those?
Answer: I tried running for 15 epochs. but it was taking a lot of time to train.So, I reduced the no. of epochs to 10 but my GPU workspace used to disconect/sleep (the limit is 30 mins of inactivity). So i used the workspace_utilities.py file provided. But since it was still taking a lot of time to train I finally reduced no. of epochs to 5. Which was good enough for a Loss of less than "3.5".
I arbitrarily chose 150 as my sequence length initially but loss started with huge number. Then I thought if in a plot of TV Script a normal conversation would be minimum 5 words and maximum 10 words (maybe 20-25 if not a monologue? ) .So, I chose 10 as my sequence length this time. Although, model's training speed was slow but the loss started with very less number(5.11) and I reached objective of less than 3.5 ( EVEN BEFORE 5th Epoch!! ) .
About the hidden dim I choose 512. Its a standard practice, we need to choose values in powers of 2 (i.e. 64,128,256,512) etc. More the value better the training.
About the n_layers, as we are using the LSTM cells its standard practice to use between 1-3 layers as we go more deeper there will high computational complexity but this is reverse in case of CNN. I used 2 as n_layers. Moreover I wanted to Use Dropout & Because of Dropout Constraints n_layers must be >= 2
As expected, We get better results with larger hidden and n_layer dimensions, but larger models take a longer time to train
After running the above training cell, your model will be saved by name, trained_rnn, and if you save your notebook progress, you can pause here and come back to this code at another time. You can resume your progress by running the next cell, which will load in our word:id dictionaries and load in your saved model by name!
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import torch
import helper
import problem_unittests as tests
_, vocab_to_int, int_to_vocab, token_dict = helper.load_preprocess()
trained_rnn = helper.load_model('./save/trained_rnn')With the network trained and saved, you'll use it to generate a new, "fake" Seinfeld TV script in this section.
To generate the text, the network needs to start with a single word and repeat its predictions until it reaches a set length. You'll be using the generate function to do this. It takes a word id to start with, prime_id, and generates a set length of text, predict_len. Also note that it uses topk sampling to introduce some randomness in choosing the most likely next word, given an output set of word scores!
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
import torch.nn.functional as F
def generate(rnn, prime_id, int_to_vocab, token_dict, pad_value, predict_len=100):
"""
Generate text using the neural network
:param decoder: The PyTorch Module that holds the trained neural network
:param prime_id: The word id to start the first prediction
:param int_to_vocab: Dict of word id keys to word values
:param token_dict: Dict of puncuation tokens keys to puncuation values
:param pad_value: The value used to pad a sequence
:param predict_len: The length of text to generate
:return: The generated text
"""
rnn.eval()
# create a sequence (batch_size=1) with the prime_id
current_seq = np.full((1, sequence_length), pad_value)
current_seq[-1][-1] = prime_id
predicted = [int_to_vocab[prime_id]]
for _ in range(predict_len):
if train_on_gpu:
current_seq = torch.LongTensor(current_seq).cuda()
else:
current_seq = torch.LongTensor(current_seq)
# initialize the hidden state
hidden = rnn.init_hidden(current_seq.size(0))
# get the output of the rnn
output, _ = rnn(current_seq, hidden)
# get the next word probabilities
p = F.softmax(output, dim=1).data
if(train_on_gpu):
p = p.cpu() # move to cpu
# use top_k sampling to get the index of the next word
top_k = 5
p, top_i = p.topk(top_k)
top_i = top_i.numpy().squeeze()
# select the likely next word index with some element of randomness
p = p.numpy().squeeze()
word_i = np.random.choice(top_i, p=p/p.sum())
# retrieve that word from the dictionary
word = int_to_vocab[word_i]
predicted.append(word)
# the generated word becomes the next "current sequence" and the cycle can continue
current_seq = np.roll(current_seq, -1, 1)
current_seq[-1][-1] = word_i
gen_sentences = ' '.join(predicted)
# Replace punctuation tokens
for key, token in token_dict.items():
ending = ' ' if key in ['\n', '(', '"'] else ''
gen_sentences = gen_sentences.replace(' ' + token.lower(), key)
gen_sentences = gen_sentences.replace('\n ', '\n')
gen_sentences = gen_sentences.replace('( ', '(')
# return all the sentences
return gen_sentencesIt's time to generate the text. Set gen_length to the length of TV script you want to generate and set prime_word to one of the following to start the prediction:
- "jerry"
- "elaine"
- "george"
- "kramer"
You can set the prime word to any word in our dictionary, but it's best to start with a name for generating a TV script. (You can also start with any other names you find in the original text file!)
# run the cell multiple times to get different results!
gen_length = 400 # modify the length to your preference
prime_word = 'jerry' # name for starting the script
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
pad_word = helper.SPECIAL_WORDS['PADDING']
generated_script = generate(trained_rnn, vocab_to_int[prime_word + ':'], int_to_vocab, token_dict, vocab_to_int[pad_word], gen_length)
print(generated_script)/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:43: UserWarning: RNN module weights are not part of single contiguous chunk of memory. This means they need to be compacted at every call, possibly greatly increasing memory usage. To compact weights again call flatten_parameters().
jerry: i'm going to the doctor!
jerry: what?
george: i was in my office for the last year, he could be able to help me out.
jerry: well, i just got to get my car back. i'm not gonna be a little more.
jerry: oh...
george: oh, yeah. i don't know what to do. i mean, you know, i know, uh, uh, i think you can take a cab.
kramer:(to jerry) what are we gonna do? i mean, i was just trying to get out of here, but, we have to have a little more time. i don't have any money to you.
jerry: i don't think so.
george: you know what, maybe you should take some of those stupid item.
jerry: oh, no. no. no.
kramer: oh, hey.
elaine:(to jerry) i know you know.
george: well, it's not the same idea, i was hoping to get some kind of a big salad.
george: i don't know what i'm gonna do for you.
elaine: i don't know, but i don't know what to say, but i don't know, but i know.
jerry: you know, the guy with a man in my house. i can't go to the hospital.
jerry: i don't know.
jerry: i don't know, i was just wondering if i could see a man.
elaine: oh, yeah, yeah, yeah, yeah.
kramer: hey, i know how to say that you could be able to find me.
elaine: i think that's the same thing. it's like a prisoner, i was hoping that i was just a big fan.
jerry: oh, no, no.
jerry:(to jerry) i know, i think you're gonna be here.(jerry gives him a look at his head. jerry turns
Once you have a script that you like (or find interesting), save it to a text file!
# save script to a text file
f = open("generated_script_6.txt","w")
f.write(generated_script)
f.close()It's ok if the TV script doesn't make perfect sense. It should look like alternating lines of dialogue, here is one such example of a few generated lines.
jerry: what about me?
jerry: i don't have to wait.
kramer:(to the sales table)
elaine:(to jerry) hey, look at this, i'm a good doctor.
newman:(to elaine) you think i have no idea of this...
elaine: oh, you better take the phone, and he was a little nervous.
kramer:(to the phone) hey, hey, jerry, i don't want to be a little bit.(to kramer and jerry) you can't.
jerry: oh, yeah. i don't even know, i know.
jerry:(to the phone) oh, i know.
kramer:(laughing) you know...(to jerry) you don't know.
You can see that there are multiple characters that say (somewhat) complete sentences, but it doesn't have to be perfect! It takes quite a while to get good results, and often, you'll have to use a smaller vocabulary (and discard uncommon words), or get more data. The Seinfeld dataset is about 3.4 MB, which is big enough for our purposes; for script generation you'll want more than 1 MB of text, generally.
When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_tv_script_generation.ipynb" and save another copy as an HTML file by clicking "File" -> "Download as.."->"html". Include the "helper.py" and "problem_unittests.py" files in your submission. Once you download these files, compress them into one zip file for submission.