How to Develop a Character-Based Neural Language Model in Keras

A language model predicts the next word in the sequence based on the specific words that have come before it in the sequence.

It is also possible to develop language models at the character level using neural networks. The benefit of character-based language models is their small vocabulary and flexibility in handling any words, punctuation, and other document structure. This comes at the cost of requiring larger models that are slower to train.

Nevertheless, in the field of neural language models, character-based models offer a lot of promise for a general, flexible and powerful approach to language modeling.

In this tutorial, you will discover how to develop a character-based neural language model.

After completing this tutorial, you will know:

  • How to prepare text for character-based language modeling.
  • How to develop a character-based language model using LSTMs.
  • How to use a trained character-based language model to generate text.

Let’s get started.

  • Update Feb/2018: Minor update to generation for API change in Keras 2.1.3.
How to Develop a Character-Based Neural Language Model in Keras

How to Develop a Character-Based Neural Language Model in Keras
Photo by hedera.baltica, some rights reserved.

Tutorial Overview

This tutorial is divided into 4 parts; they are:

  1. Sing a Song of Sixpence
  2. Data Preparation
  3. Train Language Model
  4. Generate Text

Need help with Deep Learning for Text Data?

Take my free 7-day email crash course now (with code).

Click to sign-up and also get a free PDF Ebook version of the course.

Start Your FREE Crash-Course Now

Sing a Song of Sixpence

The nursery rhyme “Sing a Song of Sixpence” is well known in the west.

The first verse is common, but there is also a 4 verse version that we will use to develop our character-based language model.

It is short, so fitting the model will be fast, but not so short that we won’t see anything interesting.

The complete 4 verse version we will use as source text is listed below.

Copy the text and save it in a new file in your current working directory with the file name ‘rhyme.txt‘.

Data Preparation

The first step is to prepare the text data.

We will start by defining the type of language model.

Language Model Design

A language model must be trained on the text, and in the case of a character-based language model, the input and output sequences must be characters.

The number of characters used as input will also define the number of characters that will need to be provided to the model in order to elicit the first predicted character.

After the first character has been generated, it can be appended to the input sequence and used as input for the model to generate the next character.

Longer sequences offer more context for the model to learn what character to output next but take longer to train and impose more burden on seeding the model when generating text.

We will use an arbitrary length of 10 characters for this model.

There is not a lot of text, and 10 characters is a few words.

We can now transform the raw text into a form that our model can learn; specifically, input and output sequences of characters.

Load Text

We must load the text into memory so that we can work with it.

Below is a function named load_doc() that will load a text file given a filename and return the loaded text.

We can call this function with the filename of the nursery rhyme ‘rhyme.txt‘ to load the text into memory. The contents of the file are then printed to screen as a sanity check.

Clean Text

Next, we need to clean the loaded text.

We will not do much to it here. Specifically, we will strip all of the new line characters so that we have one long sequence of characters separated only by white space.

You may want to explore other methods for data cleaning, such as normalizing the case to lowercase or removing punctuation in an effort to reduce the final vocabulary size and develop a smaller and leaner model.

Create Sequences

Now that we have a long list of characters, we can create our input-output sequences used to train the model.

Each input sequence will be 10 characters with one output character, making each sequence 11 characters long.

We can create the sequences by enumerating the characters in the text, starting at the 11th character at index 10.

Running this snippet, we can see that we end up with just under 400 sequences of characters for training our language model.

Save Sequences

Finally, we can save the prepared data to file so that we can load it later when we develop our model.

Below is a function save_doc() that, given a list of strings and a filename, will save the strings to file, one per line.

We can call this function and save our prepared sequences to the filename ‘char_sequences.txt‘ in our current working directory.

Complete Example

Tying all of this together, the complete code listing is provided below.

Run the example to create the ‘char_seqiences.txt‘ file.

Take a look inside you should see something like the following:

We are now ready to train our character-based neural language model.

Train Language Model

In this section, we will develop a neural language model for the prepared sequence data.

The model will read encoded characters and predict the next character in the sequence. A Long Short-Term Memory recurrent neural network hidden layer will be used to learn the context from the input sequence in order to make the predictions.

Load Data

The first step is to load the prepared character sequence data from ‘char_sequences.txt‘.

We can use the same load_doc() function developed in the previous section. Once loaded, we split the text by new line to give a list of sequences ready to be encoded.

Encode Sequences

The sequences of characters must be encoded as integers.

This means that each unique character will be assigned a specific integer value and each sequence of characters will be encoded as a sequence of integers.

We can create the mapping given a sorted set of unique characters in the raw input data. The mapping is a dictionary of character values to integer values.

Next, we can process each sequence of characters one at a time and use the dictionary mapping to look up the integer value for each character.

The result is a list of integer lists.

We need to know the size of the vocabulary later. We can retrieve this as the size of the dictionary mapping.

Running this piece, we can see that there are 38 unique characters in the input sequence data.

Split Inputs and Output

Now that the sequences have been integer encoded, we can separate the columns into input and output sequences of characters.

We can do this using a simple array slice.

Next, we need to one hot encode each character. That is, each character becomes a vector as long as the vocabulary (38 elements) with a 1 marked for the specific character. This provides a more precise input representation for the network. It also provides a clear objective for the network to predict, where a probability distribution over characters can be output by the model and compared to the ideal case of all 0 values with a 1 for the actual next character.

We can use the to_categorical() function in the Keras API to one hot encode the input and output sequences.

We are now ready to fit the model.

Fit Model

The model is defined with an input layer that takes sequences that have 10 time steps and 38 features for the one hot encoded input sequences.

Rather than specify these numbers, we use the second and third dimensions on the X input data. This is so that if we change the length of the sequences or size of the vocabulary, we do not need to change the model definition.

The model has a single LSTM hidden layer with 75 memory cells, chosen with a little trial and error.

The model has a fully connected output layer that outputs one vector with a probability distribution across all characters in the vocabulary. A softmax activation function is used on the output layer to ensure the output has the properties of a probability distribution.

Running this prints a summary of the defined network as a sanity check.

The model is learning a multi-class classification problem, therefore we use the categorical log loss intended for this type of problem. The efficient Adam implementation of gradient descent is used to optimize the model and accuracy is reported at the end of each batch update.

The model is fit for 100 training epochs, again found with a little trial and error.

Save Model

After the model is fit, we save it to file for later use.

The Keras model API provides the save() function that we can use to save the model to a single file, including weights and topology information.

We also save the mapping from characters to integers that we will need to encode any input when using the model and decode any output from the model.

Complete Example

Tying all of this together, the complete code listing for fitting the character-based neural language model is listed below.

Running the example might take one minute.

You will see that the model learns the problem well, perhaps too well for generating surprising sequences of characters.

At the end of the run, you will have two files saved to the current working directory, specifically model.h5 and mapping.pkl.

Next, we can look at using the learned model.

Generate Text

We will use the learned language model to generate new sequences of text that have the same statistical properties.

Load Model

The first step is to load the model saved to the file ‘model.h5‘.

We can use the load_model() function from the Keras API.

We also need to load the pickled dictionary for mapping characters to integers from the file ‘mapping.pkl‘. We will use the Pickle API to load the object.

We are now ready to use the loaded model.

Generate Characters

We must provide sequences of 10 characters as input to the model in order to start the generation process. We will pick these manually.

A given input sequence will need to be prepared in the same way as preparing the training data for the model.

First, the sequence of characters must be integer encoded using the loaded mapping.

Next, the sequences need to be one hot encoded using the to_categorical() Keras function.

We can then use the model to predict the next character in the sequence.

We use predict_classes() instead of predict() to directly select the integer for the character with the highest probability instead of getting the full probability distribution across the entire set of characters.

We can then decode this integer by looking up the mapping to see the character to which it maps.

This character can then be added to the input sequence. We then need to make sure that the input sequence is 10 characters by truncating the first character from the input sequence text.

We can use the pad_sequences() function from the Keras API that can perform this truncation operation.

Putting all of this together, we can define a new function named generate_seq() for using the loaded model to generate new sequences of text.

Complete Example

Tying all of this together, the complete example for generating text using the fit neural language model is listed below.

Running the example generates three sequences of text.

The first is a test to see how the model does at starting from the beginning of the rhyme. The second is a test to see how well it does at beginning in the middle of a line. The final example is a test to see how well it does with a sequence of characters never seen before.

We can see that the model did very well with the first two examples, as we would expect. We can also see that the model still generated something for the new text, but it is nonsense.

Extensions

This section lists some ideas for extending the tutorial that you may wish to explore.

  • Padding. Update the example to provides sequences line by line only and use padding to fill out each sequence to the maximum line length.
  • Sequence Length. Experiment with different sequence lengths and see how they impact the behavior of the model.
  • Tune Model. Experiment with different model configurations, such as the number of memory cells and epochs, and try to develop a better model for fewer resources.

Further Reading

This section provides more resources on the topic if you are looking go deeper.

Summary

In this tutorial, you discovered how to develop a character-based neural language model.

Specifically, you learned:

  • How to prepare text for character-based language modeling.
  • How to develop a character-based language model using LSTMs.
  • How to use a trained character-based language model to generate text.

Do you have any questions?
Ask your questions in the comments below and I will do my best to answer.


Develop Deep Learning models for Text Data Today!

Deep Learning for Natural Language Processing

Develop Your Own Text models in Minutes

…with just a few lines of python code

Discover how in my new Ebook:
Deep Learning for Natural Language Processing

It provides self-study tutorials on topics like:
Bag-of-Words, Word Embedding, Language Models, Caption Generation, Text Translation and much more…

Finally Bring Deep Learning to your Natural Language Processing Projects

Skip the Academics. Just Results.

Click to learn more.


41 Responses to How to Develop a Character-Based Neural Language Model in Keras

  1. Prakash November 7, 2017 at 12:33 am #

    Hi Jason – Thanks for sharing this article. I am in learning phase and when I try to run your program (defining the load_doc function), I am getting error. Is there any package that I need to install before I run the code ?

    Running the first set of lines for loading the doc into memory gives me the following error

    > return text
    Error: unexpected symbol in ” return text”

    • Jason Brownlee November 7, 2017 at 9:51 am #

      It looks like a copy-paste error, ensure you maintain the indenting of the Python code.

  2. Klaas November 8, 2017 at 6:54 am #

    Thanks a lot Jason. One general question. On your blog I read a lot about one hot encoding. From the mnist dataset I get it that it is easy to compare probabilities (e.g if the Number is 2 I want my network to output a 1 on the 3rd row). But when it comes to language huge vocabularies is a one hot encoding not completely inefficient? I mean e.g. 1 Million vocab size and each word a vector with one 1 and 999.999 zeros? I do not really get that.

    • Jason Brownlee November 8, 2017 at 9:31 am #

      Yep, in general we try to reduce the size of the vocab to ensure the model trains in a reasonable time.

      But what is the alternative? Perhaps less crisp word predictions and worse skill?

  3. Ravi Annaswamy November 10, 2017 at 6:47 am #

    Yet another excellent tutorial, Dr.Jason.

  4. srihari November 10, 2017 at 3:09 pm #

    Hi,

    Can we implement using nltk as helping library to keras, in transforming the text.

    Srihari

  5. Stuart November 11, 2017 at 6:24 am #

    Awesome article. Really appreciate the level of line-by-line detail.

    I think there are some mistakes around this part of the article:

    “Next, the integers need to be one hot encoded using the pad_sequences() Keras function.”

    I think you mean truncated instead of one hot encoded? Also it’s missing the accompanying code snippet.

    • Jason Brownlee November 11, 2017 at 9:28 am #

      Thanks, fixed. I meant the to_categorical() function for one hot encoding.

  6. Antonio November 24, 2017 at 7:40 am #

    Cool, thanks for sharing!

    • Jason Brownlee November 24, 2017 at 9:52 am #

      You’re welcome.

      • Antonio November 24, 2017 at 7:45 pm #

        Quick question, if I may.. if we want to characterize the sequence with some extra input features, how to we prepare the data? Just to illustrate, sticking to the above example, for example I may want to associate to each sequence the name of the Person writing the sequence. This feature may change slightly the prediction of the next character. I thought one option would be to create and train a different model for each Person, but I think this would be quite suboptimal, since majority of the rules learned will be in common to every Person and the data set will be reduced for each model/Person. Another option would be to encode the Person in the input, within each time step. But in this case there is a little bit of redundancy, since the Person input feature will be the same across all the time steps. So, is there a way to provide as input to the model a feature which is independent and unchanged across all the time steps, but which characterize the entire input sequence? Thanks very much

  7. Ethan B January 10, 2018 at 5:05 pm #

    Hi Jason, thanks for the great article!

    I have one question regarding the training phase. I was thinking about using character embeddings, for example fitting a word2vec model on characters which I would then use to train the LSTM, rather than using the one hot encoded characters. Do you think this would give any sort of performance gain? I was going to test this idea myself, but I was curious if you had tried this yourself first, or if you think it is a worthwhile approach.

  8. Al February 2, 2018 at 10:52 am #

    Hi, It’s really a nice tutorial!!
    I have one trouble. When I try to predict using generate_seq, I got this error ValueError: Error when checking : expected lstm_1_input to have shape (408, 37) but got array with shape (10, 37)
    why this would happen? Thanks!!!

    • Riya John February 28, 2018 at 1:54 am #

      Hi, I got a similar error: ValueError: cannot reshape array of size 380 into shape (1,1,10)

      Code worked for me when I commented line: encoded = encoded.reshape(1, encoded.shape[0], encoded.shape[1]) in generate_seq()

      • Franco Arda June 5, 2018 at 4:23 pm #

        @Riiya, thanks for pointing that out.
        @Jason, AI and Riya are right. Code doesn’t run (your blog or book). We need to

        #encoded = encoded.reshape(1, encoded.shape[0], encoded.shape[1])

        to make prediction work.

        • Jason Brownlee June 6, 2018 at 6:36 am #

          Are you able to confirm that you have the latest version of Keras and other libraries installed?

          • Franco Arda June 7, 2018 at 3:32 pm #

            Update now, but issue remains.
            Don’t answer for me – I’m happy that it runs.

          • Kay-Michael Würzner June 21, 2018 at 5:36 pm #

            There is definitely a problem with that reshape line:


            ValueError: cannot reshape array of size 780 into shape (1,1,10)

            Checked the versions of installed libraries. As far as I can see, keras, numpy and tensorflow are needed here.


            numpy==1.14.5
            Keras==2.2.0
            Keras-Applications==1.0.2
            Keras-Preprocessing==1.0.1
            tensorflow==1.8.0

            Commenting the reshape code indeed helps successfully running the code but given the output it looks like it screws up the actual prediction… Any ideas?

          • Jason Brownlee June 22, 2018 at 6:02 am #

            I wonder if is an issue with the data file, are you able to confirm the raw data for the text matches the post?

    • Franco Arda June 5, 2018 at 4:20 pm #

      indeed, there’s an error ….see below

  9. Neeraj April 7, 2018 at 2:26 pm #

    Hi Jason,

    Can you please help me with an error. I am new at python so many times I dont know how to resolve an error.

    Error 1 :

    mapping = load(open(‘mapping.pkl’, ‘rb’))
    Traceback (most recent call last):

    File “”, line 1, in
    mapping = load(open(‘mapping.pkl’, ‘rb’))

    NameError: name ‘load’ is not defined

    Error 2 :
    from pickle import load

    mapping = load(open(‘mapping.pkl’, ‘rb’))
    Traceback (most recent call last):

    File “”, line 1, in
    mapping = load(open(‘mapping.pkl’, ‘rb’))

    EOFError: Ran out of input

    Thanks,
    Neeraj

    • Jason Brownlee April 8, 2018 at 6:10 am #

      Perhaps double check that you have coped all of the code from the example?

  10. sagar June 2, 2018 at 1:23 am #

    Hi Jason, this is very helpful and nicely detailed. Thank you for sharing.

    I am working on a problem where I have some 32000 rows of jumbled characters “wewlsfnskfddsl…eredsda” and each row is of length 406. These are hashed, probably. And I need to predict to which class do they belong to? Here class is 1-12 names of books.

    Any suggestions on how I could modify your code above. Would my problem still need text generation? As this is a multi-class classification problem.

    Thank you very much. Looking forward to your advice.

    • Jason Brownlee June 2, 2018 at 6:38 am #

      Sounds like classification. A language mode/text generation would not be helpful.

      I would recommend testing an MLP, CNN and LSTM on the problem. Also look at some of the tutorials on the blog for sentiment classification, they will provide a template. No need for a word embedding either.

    • Koushik June 4, 2018 at 7:52 pm #

      I think many to one RNN model should work for your problem

  11. SM June 4, 2018 at 6:53 pm #

    Hi Jason, thank you for the above suggestions. I am trying to implement a LSTM model and following this post on how to set up. I have a questions about input_shape for LSTM and for dense layer.

    My xtrain is a sequence of numbers as a NumPy array. To give you a background: I have 32514 rows of jumbled characters “wewlsfnskfddsl…” , which I reshaped: X = X.reshape((1,32514,1)) into 3D for in LSTM’s input_shape, taking inspiration from your post “https://machinelearningmastery.com/reshape-input-data-long-short-term-memory-networks-keras/”.

    However, here my y is of 12 classes (0 to 11). It is a multiclass classification problem.

    y = to_categorical(y, num_classes=12)

    How should I define my dense layer? According to keras document, here’s what the input_shape is for dense and LSTM layers:

    Dense: (batch_size, input_dim)
    LSTM: (batch_size, timesteps, input_dim)

    # define model
    model = Sequential()
    model.add(LSTM(75, input_shape=(32514,1)))
    model.add(Dense(input(1), activation=’softmax’))
    print(model.summary())
    # compile model
    model.compile(loss=’categorical_crossentropy’, optimizer=’adam’, metrics=[‘accuracy’])
    # fit model
    model.fit(X, y, epochs=100, verbose=2)

    # save the model to file
    model.save(‘model.h5’)
    # save the mapping
    dump(mapping, open(‘mapping.pkl’, ‘wb’))

    When I run the above cell, the program seems to be asking to for another input.

    Is this correct? Thank you very much. I love your blog.

  12. Ahmed Sahlol June 11, 2018 at 12:53 am #

    Many thanks Jason for the topic and the detailed explanation, really looks awesome.
    I have a question: Why u assigned the LSTM hidden layer = 75 memory cells?

    • Jason Brownlee June 11, 2018 at 6:09 am #

      The model was configured with a little trial and error.

  13. Christian June 27, 2018 at 12:14 am #

    Hi Jason,

    Do u know if I can use this technique (character) to cluster rare words, like DriverId with Driver, Name, etc ?. Or Vehicle with vehicleId, location, latitude, longitude?,

    Thanks.

    • Jason Brownlee June 27, 2018 at 8:19 am #

      Perhaps you can use an embedding for these features?

  14. Ayden M. June 28, 2018 at 10:11 pm #

    Thank you so much for this tutorial. I am getting this error when training my LM, and I can’t figure out how to overcome this:

    IndexError: too many indices for array

    It occurs in line 39: X, y = sequences[:,:-1], sequences[:,-1]

    From what I understand, it doesn’t want to create a 2D vector from a 1D vector. I tried to reshape with numpy but I keep getting errors of similar nature. Do you have any idea how to solve this?

  15. Tuomas V. September 25, 2018 at 12:33 am #

    Hello!

    Thanks for the incredibly easy-to-follow tutorial.

    I’m wondering whether there’s a more memory-friendly way to one-hot encode than using to_categorical?

    The sequences text file I’m working with is nearly 1 Gb in size, my vocabulary size is quite large and thus there are MemoryErrors at:

    sequences = [to_categorical(x, num_classes=vocab_size) for x in X]

    Any help would be appreciated!

    • Jason Brownlee September 25, 2018 at 6:25 am #

      Hmmm. Some thoughts:

      Don’t use one hot encoding, use integer encoding and a word embedding and process the file in chunks via progressive loading.

      • Tuomas V. October 10, 2018 at 10:41 pm #

        Droppped one-hot encoding, switched to fastText word vectors and it has been smooth sailing since. Thanks!

Leave a Reply