SALE! Use code blackfriday for 40% off everything!
Hurry, sale ends soon! Click to see the full catalog.

Understanding Simple Recurrent Neural Networks in Keras

This tutorial is designed for anyone looking for an understanding of how recurrent neural networks (RNN) work and how to use them via the Keras deep learning library. While the Keras library provides all the methods required for solving problems and building applications, it is also important to gain an insight into how everything works. In this article, the computations taking place in the RNN model are shown step by step. Next, a complete end-to-end system for time series prediction is developed.

After completing this tutorial, you will know:

  • The structure of an RNN
  • How an RNN computes the output when given an input
  • How to prepare data for a SimpleRNN in Keras
  • How to train a SimpleRNN model

Kick-start your project with my book Building Transformer Models with Attention. It provides self-study tutorials with working code to guide you into building a fully-working transformer model that can
translate sentences from one language to another...

Let’s get started.

Umstead state park

Understanding simple recurrent neural networks in Keras. Photo by Mehreen Saeed, some rights reserved.

Tutorial Overview

This tutorial is divided into two parts; they are:

  1. The structure of the RNN
    1. Different weights and biases associated with different layers of the RNN
    2. How computations are performed to compute the output when given an input
  2. A complete application for time series prediction


It is assumed that you have a basic understanding of RNNs before you start implementing them. An Introduction to Recurrent Neural Networks and the Math That Powers Them gives you a quick overview of RNNs.

Let’s now get right down to the implementation part.

Import Section

To start the implementation of RNNs, let’s add the import section.

Want to Get Started With Building Transformer Models with Attention?

Take my free 12-day email crash course now (with sample code).

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

Keras SimpleRNN

The function below returns a model that includes a SimpleRNN layer and a Dense layer for learning sequential data. The input_shape specifies the parameter (time_steps x features). We’ll simplify everything and use univariate data, i.e., one feature only; the time steps are discussed below.

The object demo_model is returned with two hidden units created via the SimpleRNN layer and one dense unit created via the Dense layer. The input_shape is set at 3×1, and a linear activation function is used in both layers for simplicity. Just to recall, the linear activation function $f(x) = x$ makes no change in the input. The network looks as follows:

If we have $m$ hidden units ($m=2$ in the above case), then:

  • Input: $x \in R$
  • Hidden unit: $h \in R^m$
  • Weights for the input units: $w_x \in R^m$
  • Weights for the hidden units: $w_h \in R^{mxm}$
  • Bias for the hidden units: $b_h \in R^m$
  • Weight for the dense layer: $w_y \in R^m$
  • Bias for the dense layer: $b_y \in R$

Let’s look at the above weights. Note: As the weights are randomly initialized, the results posted here will be different from yours. The important thing is to learn what the structure of each object being used looks like and how it interacts with others to produce the final output.

Now let’s do a simple experiment to see how the layers from a SimpleRNN and Dense layer produce an output. Keep this figure in view.

Layers Of A Recurrent Neural Network

Layers of a recurrent neural network

We’ll input x for three time steps and let the network generate an output. The values of the hidden units at time steps 1, 2, and 3 will be computed. $h_0$ is initialized to the zero vector. The output $o_3$ is computed from $h_3$ and $w_y$. An activation function is not required as we are using linear units.

Running the RNN on Sunspots Dataset

Now that we understand how the SimpleRNN and Dense layers are put together. Let’s run a complete RNN on a simple time series dataset. We’ll need to follow these steps:

  1. Read the dataset from a given URL
  2. Split the data into training and test sets
  3. Prepare the input to the required Keras format
  4. Create an RNN model and train it
  5. Make the predictions on training and test sets and print the root mean square error on both sets
  6. View the result

Step 1, 2: Reading Data and Splitting Into Train and Test

The following function reads the train and test data from a given URL and splits it into a given percentage of train and test data. It returns single-dimensional arrays for train and test data after scaling the data between 0 and 1 using MinMaxScaler from scikit-learn.

Step 3: Reshaping Data for Keras

The next step is to prepare the data for Keras model training. The input array should be shaped as: total_samples x time_steps x features.

There are many ways of preparing time series data for training. We’ll create input rows with non-overlapping time steps. An example for time steps = 2 is shown in the figure below. Here, time steps denotes the number of previous time steps to use for predicting the next value of the time series data.

How Data Is Prepared For Sunspots Example

How data is prepared for sunspots example

The following function get_XY() takes a one-dimensional array as input and converts it to the required input X and target Y arrays. We’ll use 12 time_steps for the sunspots dataset as the sunspots generally have a cycle of 12 months. You can experiment with other values of time_steps.

Step 4: Create RNN Model and Train

For this step, you can reuse your create_RNN() function that was defined above.

Step 5: Compute and Print the Root Mean Square Error

The function print_error() computes the mean square error between the actual and predicted values.

Step 6: View the Result

The following function plots the actual target values and the predicted values. The red line separates the training and test data points.

The following plot is generated:

Consolidated Code

Given below is the entire code for this tutorial. Try this out at your end and experiment with different hidden units and time steps. You can add a second SimpleRNN to the network and see how it behaves. You can also use the scaler object to rescale the data back to its normal range.

Further Reading

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




In this tutorial, you discovered recurrent neural networks and their various architectures.

Specifically, you learned:

  • The structure of RNNs
  • How the RNN computes an output from previous inputs
  • How to implement an end-to-end system for time series forecasting using an RNN

Do you have any questions about RNNs discussed in this post? Ask your questions in the comments below, and I will do my best to answer.

Learn Transformers and Attention!

Building Transformer Models with Attention

Teach your deep learning model to read a sentence

...using transformer models with attention

Discover how in my new Ebook:
Building Transformer Models with Attention

It provides self-study tutorials with working code to guide you into building a fully-working transformer models that can
translate sentences from one language to another...

Give magical power of understanding human language for
Your Projects

See What's Inside

, , , , , ,

17 Responses to Understanding Simple Recurrent Neural Networks in Keras

  1. Avatar
    tony November 15, 2021 at 8:47 pm #

    When I use tensorflow.keras, NotImplementedError is thrown in create_RNN(). What should I update? Thanks.

    • Avatar
      Adrian Tam November 16, 2021 at 2:29 am #

      create_RNN() is a function defined here. Which line is triggering the NotImplementedError?

      • Avatar
        tony November 16, 2021 at 3:12 pm #

        In “model.add(SimpleRNN(hidden_units, input_shape=input_shape, activation=activation[0]))”

        The error thrown at:
        ~/anaconda3/envs/py38/lib/python3.8/site-packages/tensorflow/python/framework/ in __array__(self)
        851 def __array__(self):
        –> 852 raise NotImplementedError(
        853 “Cannot convert a symbolic Tensor ({}) to a numpy array.”
        854 ” This error may indicate that you’re trying to pass a Tensor to”

        NotImplementedError: Cannot convert a symbolic Tensor (simple_rnn/strided_slice:0) to a numpy array. This error may indicate that you’re trying to pass a Tensor to a NumPy call, which is not supported

        Python version is 3.8.12
        Tensorflow version is 2.4.1


  2. Avatar
    Master56^^ December 21, 2021 at 6:40 am #

    Error still wasn’t fix NotImplementedError: Cannot convert a symbolic Tensor (simple_rnn_1/strided_slice:0) to a numpy array. This error may indicate that you’re trying to pass a Tensor to a NumPy call, which is not supported

    • Avatar
      James Carmichael December 24, 2021 at 5:58 am #

      Hi Master56^^…Please provide a full code listing and a description of your Python coding environment.


  3. Avatar
    Tommy Guo August 9, 2022 at 9:26 am #

    Hi Mehran,

    Thank you for providing this very insightful post! I had gone through learning materials from other online courses and labs but did not actually grasp what was actually happening in a RNN until I read your post here. It was very well written and elegantly lucid. The code is similarly very easy to follow and I hope to expand on this example to implement an RNN for my research project!

    All the best,


    • Avatar
      James Carmichael August 9, 2022 at 9:53 am #

      Great feedback Tommy! We appreciate your support and feedback!

  4. Avatar
    Paolo March 3, 2023 at 5:22 am #

    really great post!
    I might have spotted a little typo:

    h1 =[0], wx) + h0 + bh

    should be:

    h1 =[0], wx) +, wh) + bh

    anyway, the result doesn’t change as h0 is all zeros

    • Avatar
      James Carmichael March 3, 2023 at 9:23 am #

      Thank you Paolo! We appreciate the feedback.

  5. Avatar
    cesiy March 28, 2023 at 5:30 pm #

    we want extra information about this and it also execute on copy paste in colab or jupyter

    • Avatar
      James Carmichael March 29, 2023 at 7:16 am #

      Hi cesiy…Have you been able to successfully execute the sample code? If not please let us know what is not working or what error messages you are receiving.

  6. Avatar
    tasos July 10, 2023 at 8:57 pm #

    Thank you! How about a sequence with more than one feature per timestep? I tried the following but Keras outputs one value, and the manual computation outputs two values:

    # hidden_units, dense_units, input_shape, activation
    m = 2
    demo_model = create_RNN(m, 1, (3,2), activation=[‘tanh’, ‘linear’])

    wx = demo_model.get_weights()[0]
    wh = demo_model.get_weights()[1]
    bh = demo_model.get_weights()[2]
    wy = demo_model.get_weights()[3]
    by = demo_model.get_weights()[4]

    print(‘wx = ‘, wx, ‘\nwh = ‘, wh, ‘\nbh = ‘, bh, ‘\nwy =’, wy, ‘\nby = ‘, by)

    x = np.array([1,0.1, 2,0.2, 3,0.3])
    # Reshape the input to the required sample_size x time_steps x features
    x_input = np.reshape(x,(1, 3, 2))
    print(‘x =’, x)
    print(‘x_reshaped = ‘, x_input)
    y_pred_model = demo_model.predict(x_input)

    h0 = np.zeros(m)
    h1 = np.tanh([0], wx) +,wh) + bh)
    h2 = np.tanh([1], wx) +,wh) + bh)
    h3 = np.tanh([2], wx) +,wh) + bh)
    o3 =, wy) + by

    print(‘h1 = ‘, h1, ‘h2 = ‘, h2, ‘h3 = ‘, h3)

    print(“Prediction from network “, y_pred_model)
    print(“Prediction from our computation “, o3)

    Prediction from network [[0.09162921]]
    Prediction from our computation [[ 0.14343597] [-0.27000149]]

    • Avatar
      James Carmichael July 10, 2023 at 11:12 pm #

      Hi Tasos…Thank you for your feedback! We will review this scenario.

  7. Avatar
    Odise October 4, 2023 at 7:05 am #

    Hello, im using this dataset at the moment and i have to use a split ( train test validation) where train: from first row until 1965, validation from 1966 to 1975 and test from 1976 to last row. Im currently new to this type of language since i just started using Rstudio and i would like an answer as soon as posible bcuz i don’t seem to find a solution. Thank you very much.

  8. Avatar
    Zineb October 20, 2023 at 9:13 pm #

    In the first example (Keras simpleRNN), you get weights without training the model ( no fit functing after model compilation )

    how is that ?

    Thanks in advance

  9. Avatar
    Zineb October 20, 2023 at 10:42 pm #

    One more question PLZ

    In the first example (Keras Simple RNN);
    print(“Prediction from network “, y_pred_model)
    I obtain : [[-1.1634194]] !!

    I just copy and past the code you give

Leave a Reply