Applied Deep Learning in Python Mini-Course

Last Updated on December 11, 2019

Deep learning is a fascinating field of study and the techniques are achieving world class results in a range of challenging machine learning problems.

It can be hard to get started in deep learning.

Which library should you use and which techniques should you focus on?

In this post you will discover a 14-part crash course into deep learning in Python with the easy to use and powerful Keras library.

This mini-course is intended for python machine learning practitioners that are already comfortable with scikit-learn on the SciPy ecosystem for machine learning.

Kick-start your project with my new book Deep Learning With Python, including step-by-step tutorials and the Python source code files for all examples.

Let’s get started.

(Tip: you might want to print or bookmark this page so that you can refer back to it later.)

  • Update Mar/2018: Added alternate link to download the dataset as the original appears to have been taken down.
  • Update Oct/2019: Updated for Keras 2.3.0.
Applied Deep Learning in Python Mini-Course

Applied Deep Learning in Python Mini-Course
Photo by darkday, some rights reserved.

Who Is This Mini-Course For?

Before we get started, let’s make sure you are in the right place. The list below provides some general guidelines as to who this course was designed for.

Don’t panic if you don’t match these points exactly, you might just need to brush up in one area or another to keep up.

  • Developers that know how to write a little code. This means that it is not a big deal for you to get things done with Python and know how to setup the SciPy ecosystem on your workstation (a prerequisite). It does not mean your a wizard coder, but it does mean you’re not afraid to install packages and write scripts.
  • Developers that know a little machine learning. This means you know about the basics of machine learning like cross validation, some algorithms and the bias-variance trade-off. It does not mean that you are a machine learning PhD, just that you know the landmarks or know where to look them up.

This mini-course is not a textbook on Deep Learning.

It will take you from a developer that knows a little machine learning in Python to a developer who can get results and bring the power of Deep Learning to your own projects.

Mini-Course Overview (what to expect)

This mini-course is divided into 14 parts.

Each lesson was designed to take the average developer about 30 minutes. You might finish some much sooner and other you may choose to go deeper and spend more time.

You can can complete each part as quickly or as slowly as you like. A comfortable schedule may be to complete one lesson per day over a two week period. Highly recommended.

The topics you will cover over the next 14 lessons are as follows:

  • Lesson 01: Introduction to Theano.
  • Lesson 02: Introduction to TensorFlow.
  • Lesson 03: Introduction to Keras.
  • Lesson 04: Crash Course in Multi-Layer Perceptrons.
  • Lesson 05: Develop Your First Neural Network in Keras.
  • Lesson 06: Use Keras Models With Scikit-Learn.
  • Lesson 07: Plot Model Training History.
  • Lesson 08: Save Your Best Model During Training With Checkpointing.
  • Lesson 09: Reduce Overfitting With Dropout Regularization.
  • Lesson 10: Lift Performance With Learning Rate Schedules.
  • Lesson 11: Crash Course in Convolutional Neural Networks.
  • Lesson 12: Handwritten Digit Recognition.
  • Lesson 13: Object Recognition in Small Photographs.
  • Lesson 14: Improve Generalization With Data Augmentation.

This is going to be a lot of fun.

You’re going to have to do some work though, a little reading, a little research and a little programming. You want to learn deep learning right?

(Tip: All of the answers these lessons can be found on this blog, use the search feature)

Any questions at all, please post in the comments below.

Share your results in the comments.

Hang in there, don’t give up!

Need help with Deep Learning in Python?

Take my free 2-week email course and discover MLPs, CNNs and LSTMs (with code).

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

Start Your FREE Mini-Course Now!

Lesson 01: Introduction to Theano

Theano is a Python library for fast numerical computation to aid in the development of deep learning models.

At it’s heart Theano is a compiler for mathematical expressions in Python. It knows how to take your structures and turn them into very efficient code that uses NumPy and efficient native libraries to run as fast as possible on CPUs or GPUs.

The actual syntax of Theano expressions is symbolic, which can be off-putting to beginners used to normal software development. Specifically, expression are defined in the abstract sense, compiled and later actually used to make calculations.

In this lesson your goal is to install Theano and write a small example that demonstrates the symbolic nature of Theano programs.

For example, you can install Theano using pip as follows:

A small example of a Theano program that you can use as a starting point is listed below:

Learn more about Theano on the Theano homepage.

Lesson 02: Introduction to TensorFlow

TensorFlow is a Python library for fast numerical computing created and released by Google. Like Theano, TensorFlow is intended to be used to develop deep learning models.

With the backing of Google, perhaps used in some of it’s production systems and used by the Google DeepMind research group, it is a platform that we cannot ignore.

Unlike Theano, TensorFlow does have more of a production focus with a capability to run on CPUs, GPUs and even very large clusters.

In this lesson your goal is to install TensorFlow become familiar with the syntax of the symbolic expressions used in TensorFlow programs.

For example, you can install TensorFlow using pip:

A small example of a TensorFlow program that you can use as a starting point is listed below:

Learn more about TensorFlow on the TensorFlow homepage.

Lesson 03: Introduction to Keras

A difficulty of both Theano and TensorFlow is that it can take a lot of code to create even very simple neural network models.

These libraries were designed primarily as a platform for research and development more than for the practical concerns of applied deep learning.

The Keras library addresses these concerns by providing a wrapper for both Theano and TensorFlow. It provides a clean and simple API that allows you to define and evaluate deep learning models in just a few lines of code.

Because of the ease of use and because it leverages the power of Theano and TensorFlow, Keras is quickly becoming the go-to library for applied deep learning.

The focus of Keras is the concept of a model. The life-cycle of a model can be summarized as follows:

  1. Define your model. Create a Sequential model and add configured layers.
  2. Compile your model. Specify loss function and optimizers and call the compile()
    function on the model.
  3. Fit your model. Train the model on a sample of data by calling the fit() function on
    the model.
  4. Make predictions. Use the model to generate predictions on new data by calling functions such as evaluate() or predict() on the model.

Your goal for this lesson is to install Keras.

For example, you can install Keras using pip:

Start to familiarize yourself with the Keras library ready for the upcoming lessons where we will implement our first model.

You can learn more about the Keras library on the Keras homepage.

Lesson 04: Crash Course in Multi-Layer Perceptrons

Artificial neural networks are a fascinating area of study, although they can be intimidating
when just getting started.

The field of artificial neural networks is often just called neural networks or multi-layer
Perceptrons after perhaps the most useful type of neural network.

The building block for neural networks are artificial neurons. These are simple computational
units that have weighted input signals and produce an output signal using an activation function.

Neurons are arranged into networks of neurons. A row of neurons is called a layer and one
network can have multiple layers. The architecture of the neurons in the network is often called the network topology.

Once configured, the neural network needs to be trained on your dataset. The classical and still preferred training algorithm for neural networks is called stochastic
gradient descent.

Model of a Simple Neuron

Model of a Simple Neuron

Your goal for this lesson is to become familiar with neural network terminology.

Dig a little deeper into terms like neuron, weights, activation function, learning rate and more.

Lesson 05: Develop Your First Neural Network in Keras

Keras allows you to develop and evaluate deep learning models in very few lines of code.

In this lesson your goal is to develop your first neural network using the Keras library.

Use a standard binary (two-class) classification dataset from the UCI Machine Learning Repository, like the Pima Indians onset of diabetes or the ionosphere datasets.

Piece together code to achieve the following:

  1. Load your dataset using NumPy or Pandas.
  2. Define your neural network model and compile it.
  3. Fit your model to the dataset.
  4. Estimate the performance of your model on unseen data.

To give you a massive kick start, below is a complete working example that you can use as a starting point.

Download the dataset and place it in your current working directory.

Now develop your own model on a different dataset, or adapt this example.

Learn more about the Keras API for simple model development.

Lesson 06: Use Keras Models With Scikit-Learn

The scikit-learn library is a general purpose machine learning framework in Python built on top of SciPy.

Scikit-learn excels at tasks such as evaluating model performance and optimizing model hyperparameters in just a few lines of code.

Keras provides a wrapper class that allows you to use your deep learning models with scikit-learn. For example, an instance of KerasClassifier class in Keras can wrap your deep learning model and be used as an Estimator in scikit-learn.

When using the KerasClassifier class, you must specify the name of a function that the class can use to define and compile your model. You can also pass additional parameters to the constructor of the KerasClassifier class that will be passed to the call later, like the number of epochs and batch size.

In this lesson your goal is to develop a deep learning model and evaluate it using k-fold cross validation.

For example, you can define an instance of the KerasClassifier and the custom function to create your model as follows:

Learn more about using your Keras deep learning models with scikit-learn on the Wrappers for the Sciki-Learn API webpage.

Lesson 07: Plot Model Training History

You can learn a lot about neural networks and deep learning models by observing their performance over time during training.

Keras provides the capability to register callbacks when training a deep learning model.

One of the default callbacks that is registered when training all deep learning models is the History callback. It records training metrics for each epoch. This includes the loss and the accuracy (for classification problems) as well as the loss and accuracy for the validation dataset, if one is set.

The history object is returned from calls to the fit() function used to train the model. Metrics are stored in a dictionary in the history member of the object returned.

Your goal for this lesson is to investigate the history object and create plots of model performance during training.

For example, you can print the list of metrics collected by your history object as follows:

You can learn more about the History object and the callback API in Keras.

Lesson 08: Save Your Best Model During Training With Checkpointing

Application checkpointing is a fault tolerance technique for long running processes.

The Keras library provides a checkpointing capability by a callback API. The ModelCheckpoint
callback class allows you to define where to checkpoint the model weights, how the file should
be named and under what circumstances to make a checkpoint of the model.

Checkpointing can be useful to keep track of the model weights in case your training run is stopped prematurely. It is also useful to keep track of the best model observed during training.

In this lesson, your goal is to use the ModelCheckpoint callback in Keras to keep track of the best model observed during training.

You could define a ModelCheckpoint that saves network weights to the same file each time an improvement is observed. For example:

Learn more about using the ModelCheckpoint callback in Keras.

Lesson 09: Reduce Overfitting With Dropout Regularization

A big problem with neural networks is that they can overlearn your training dataset.

Dropout is a simple yet very effective technique for reducing dropout and has proven useful in large deep learning models.

Dropout is a technique where randomly selected neurons are ignored during training. They are dropped-out randomly. This means that their contribution to the activation of downstream neurons is temporally removed on the forward pass and any weight updates are not applied to the neuron on the backward pass.

You can add a dropout layer to your deep learning model using the Dropout layer class.

In this lesson your goal is to experiment with adding dropout at different points in your neural network and set to different probability of dropout values.

For example, you can create a dropout layer with the probability of 20% and add it to your model as follows:

You can learn more about dropout in Keras.

Lesson 10: Lift Performance With Learning Rate Schedules

You can often get a boost in the performance of your model by using a learning rate schedule.

Often called an adaptive learning rate or an annealed learning rate, this is a technique where the learning rate used by stochastic gradient descent changes while training your model.

Keras has a time-based learning rate schedule built into the implementation of the stochastic gradient descent algorithm in the SGD class.

When constructing the class, you can specify the decay which is the amount that your learning rate (also specified) will decrease each epoch. When using learning rate decay you should bump up your initial learning rate and consider adding a large momentum value such as 0.8 or 0.9.

Your goal in this lesson is to experiment with the time-based learning rate schedule built into Keras.

For example, you can specify a learning rate schedule that starts at 0.1 and drops by 0.0001 each epoch as follows:

You can learn more about the SGD class in Keras here.

Lesson 11: Crash Course in Convolutional Neural Networks

Convolutional Neural Networks are a powerful artificial neural network technique.

They expect and preserve the spatial relationship between pixels in images by learning internal feature representations using small squares of input data.

Feature are learned and used across the whole image, allowing for the objects in your images to be shifted or translated in the scene and still detectable by the network. It is this reason why this type of network is so useful for object recognition in photographs, picking out digits, faces, objects and so on with varying orientation.

There are three types of layers in a Convolutional Neural Network:

  1. Convolutional Layers comprised of filters and feature maps.
  2. Pooling Layers that down sample the activations from feature maps.
  3. Fully-Connected Layers that plug on the end of the model and can be used to make predictions.

In this lesson you are to familiarize yourself with the terminology used when describing convolutional neural networks.

This may require a little research on your behalf.

Don’t worry too much about how they work just yet, just learn the terminology and configuration of the various layers used in this type of network.

Lesson 12: Handwritten Digit Recognition

Handwriting digit recognition is a difficult computer vision classification problem.

The MNIST dataset is a standard problem for evaluating algorithms on the problem of handwriting digit recognition. It contains 60,000 images of digits that can be used to train a model, and 10,000 images that can be used to evaluate its performance.

Example MNIST images

Example MNIST images

State of the art results can be achieved on the MNIST problem using convolutional neural networks. Keras makes loading the MNIST dataset dead easy.

In this lesson, your goal is to develop a very simple convolutional neural network for the MNIST problem comprised of one convolutional layer, one max pooling layer and one dense layer to make predictions.

For example, you can load the MNIST dataset in Keras as follows:

It may take a moment to download the files to your computer.

As a tip, the Keras Conv2D layer that you will use as your first hidden layer expects image data in the format width x height x channels, where the MNIST data has 1 channel because the images are gray scale and a width and height of 28 pixels. You can easily reshape the MNIST dataset as follows:

You will also need to one-hot encode the output class value, that Keras also provides a handy helper function to achieve:

As a final tip, here is a model definition that you can use as a starting point:

Lesson 13: Object Recognition in Small Photographs

Object recognition is a problem where your model must indicate what is in a photograph.

Deep learning models achieve state of the art results in this problem using deep convolutional neural networks.

A popular standard dataset for evaluating models on this type of problem is called CIFAR-10. It contains 60,000 small photographs, each of one of 10 objects, like a cat, ship or airplane.

Small Sample of CIFAR-10 Images

Small Sample of CIFAR-10 Images

As with the MNIST dataset, Keras provides a convenient function that you can use to load the dataset, and it will download it to your computer the first time you try to load it. The dataset is a 163 MB so it may take a few minutes to download.

Your goal in this lesson is to develop a deep convolutional neural network for the CIFAR-10 dataset. I would recommend a repeated pattern of convolution and pooling layers. Consider experimenting with drop-out and long training times.

For example, you can load the CIFAR-10 dataset in Keras and prepare it for use with a convolutional neural network as follows:

Lesson 14: Improve Generalization With Data Augmentation

Data preparation is required when working with neural network and deep learning models.

Increasingly data augmentation is also required on more complex object recognition tasks. This is where images in your dataset are modified with random flips and shifts. This in essence makes your training dataset larger and helps your model to generalize the position and orientation of objects in images.

Keras provides an image augmentation API that will create modified versions of images in your dataset just-in-time. The ImageDataGenerator class can be used to define the image augmentation operations to perform which can be fit to a dataset and then used in place of your dataset when training your model.

Your goal with this lesson is to experiment with the Keras image augmentation API using a dataset you are already familiar with from a previous lesson like MNIST or CIFAR-10.

For example, the example below creates random rotations of up to 90 degrees of images in the MNIST dataset.

You can learn more about the Keras image augmentation API.

Deep Learning Mini-Course Review

Congratulations, you made it. Well done!

Take a moment and look back at how far you have come:

  • You discovered deep learning libraries in python including the powerful numerical libraries Theano and TensorFlow and the easy to use Keras library for applied deep learning.
  • You built your first neural network using Keras and learned how to use your deep learning models with scikit-learn and how to retrieve and plot the training history for your models.
  • You learned about more advanced techniques such as dropout regularization and learning rate schedules and how you can use these techniques in Keras.
  • Finally, you took the next step and learned about and developed convolutional neural networks for complex computer vision tasks and learned about augmentation of image data.

Don’t make light of this, you have come a long way in a short amount of time. This is just the beginning of your journey with deep learning in python. Keep practicing and developing your skills.

Did you enjoy this mini-course? Do you have any questions or sticking points?
Leave a comment and let me know.

Develop Deep Learning Projects with Python!

Deep Learning with Python

 What If You Could Develop A Network in Minutes

...with just a few lines of Python

Discover how in my new Ebook:
Deep Learning With Python

It covers end-to-end projects on topics like:
Multilayer PerceptronsConvolutional Nets and Recurrent Neural Nets, and more...

Finally Bring Deep Learning To
Your Own Projects

Skip the Academics. Just Results.

See What's Inside

88 Responses to Applied Deep Learning in Python Mini-Course

  1. Ranjeet Singh July 6, 2016 at 8:52 pm #

    Very less content in each lesson

    • Jason Brownlee July 7, 2016 at 7:26 am #

      Well, it is a mini course not a university course.

      • Asad April 10, 2020 at 4:56 pm #

        dear Jason.
        I highly appreciate you effort. the course is great .i have one question regarding the part which is missing in the course . it is about to use the model. i built models based on lesson 6. i trained and adjust the model . my question what function i have to use to ask model to predict the values if i have new feed data. usually model predict should work. but what this model it did not work. can you advice me! how we can do it.

  2. KARAN VERMA July 6, 2016 at 11:59 pm #

    Hi Jason, what hardware do you use ? Do you use a power laptop ?

  3. Seyed July 11, 2016 at 9:46 pm #

    Hi Dear
    Unfortunately, I could not Download the Deep Learning Mini-Course file and others. what should I do?

    • Jason Brownlee July 12, 2016 at 5:26 am #

      Sorry to hear that Seyed, normally it is emailed to you immediately. Perhaps it landed in another folder?

      I have just sent you a direct email with the PDF attached.

  4. Daniel February 2, 2018 at 4:55 pm #

    Dear Dr. Brownlee
    Why do we evaluate the model on the same training set (X,Y). Is that somewhat different from other Machine Learning models which never use training data as evaluation ?

  5. manohar palanisamy May 16, 2018 at 1:14 am #

    In Lesson 02: Introduction to TensorFlow

    result is not defined

    change result to c

  6. Rajesh Arasada May 31, 2018 at 12:37 am #


    I am working through the code in Chp.7 DeepLearning Book. I am encountering two problems.

    1. when loading data using np.loadtxt() function I encounter the following error: ValueError: could not convert string to float: ‘column_a’.

    I loaded the data instead using np.genfromtxt() function and went through the code until I encountered a second error when training the model

    2. the acc: nan

    Epoch 28/150
    768/768 [==============================] – 0s 122us/step – loss: nan – acc: 0.0000e+00
    Epoch 29/150
    768/768 [==============================] – 0s 125us/step – loss: nan – acc: 0.0000e+00
    Epoch 30/150
    768/768 [==============================] – 0s 159us/step – loss: nan – acc: 0.0000e+00

    Not sure what the problem is. Any comments where I went wrong.
    thanking you,

    • Jason Brownlee May 31, 2018 at 6:20 am #

      Perhaps you need to update your libraries?
      Perhaps you could try running the code files provided with the book?

      Let me know how you go here or via email (contact page).

  7. Mehmet September 22, 2018 at 6:17 am #

    In lesson 12 you wrote that the Conv2D expects data in the format channels x width x height. But here the channel parameter comes at last. Am i correct?

    • Jason Brownlee September 22, 2018 at 6:33 am #


      • Mehmet September 22, 2018 at 7:28 am #

        Thank you for the quick response! Furthermore i thank you so much for this site. It helps me a lot!

  8. Hschool_student July 12, 2019 at 10:30 am #

    Hello i’m a high school student trying to learn about parallel processing and computer science, how does Applied deep learning relate to Hypercubes?

    • Jason Brownlee July 13, 2019 at 6:48 am #

      Deep learning is a field of study.

      A hypercube is a concept from geometry, e.g. a cube in more than 3 dimensions.

  9. Yoni Krichevsky December 12, 2019 at 10:16 pm #

    Is there a tensorflow 2 similar material?

    • Jason Brownlee December 13, 2019 at 6:02 am #

      I mainly use standalone keras, but you can adapt all the examples to tf.keras if you like.

  10. Vaishali April 5, 2020 at 10:58 pm #

    Thank you for sending the course details.
    I have installed theano & execute the code also.
    It works.
    Thank you.

  11. Vicente April 6, 2020 at 12:06 am #

    Hi Jason, in Lesson 04 I have a problem.

    I have copied the content of the link of the pima-indians-diabetes database into an excel file and saved it as .csv (in 4 different .csv formats it allows) and when I run the code in the Anaconda prompt it always gives me an error:

    File “”, line 5, in
    dataset = numpy.loadtxt(“pima-indians-diabetes.csv”, delimiter=’,’)
    NameError: name ‘numpy’ is not defined

    I have numpy installed (I have checked it via pip install numpy). Why it says “name ‘numpy’ is not defined?

    • Jason Brownlee April 6, 2020 at 6:08 am #

      The error suggests that numpy is not installed or if it is installed it is not available at your anaconda promopt.

      From this same prompt, perhaps try installing it again, then reboot the machine?

  12. Asad April 10, 2020 at 6:20 pm #

    Dear Jason ,
    First I would like to thank you for the online course.
    I have question , I have built an model based on lesson 6 , i trained with data etc . Now I want to use it to predict the values for new set of data , which function I should use ,
    I tried with model predict , it did not work .
    Here is the script
    from keras.models import Sequential
    from keras.layers import Dense
    from keras.wrappers.scikit_learn import KerasRegressor
    from sklearn.model_selection import cross_val_score
    from sklearn.model_selection import KFold
    from sklearn.preprocessing import StandardScaler
    from sklearn.pipeline import Pipeline
    import numpy
    import timeit
    import pandas as pd
    start = timeit.default_timer()
    names = [‘WOB’, ‘RPM’, ‘ROP’, ‘SPP’,’Flow’ ,’Flow2′]
    inputdata = pd.read_csv(“Input5.csv”, names=names)
    dataset = inputdata.values
    X = dataset[:,0:5]
    Y = dataset[:,5]
    test_data= dataset[:,0:5]
    seed = None
    def baseline_model():
    # create model
    model = Sequential()
    model.add(Dense(10, input_dim=5, kernel_initializer=’normal’, activation=’relu’))
    model.add(Dense(10, kernel_initializer=’normal’, activation=’relu’))
    model.add(Dense(1, kernel_initializer=’normal’))
    # Compile model
    model.compile(loss=’mean_squared_error’, optimizer=’adam’)
    return model
    estimator = KerasRegressor(build_fn=baseline_model, nb_epoch=100, batch_size=10, verbose=0)
    kfold = KFold(n_splits=2, random_state=seed)
    results = cross_val_score(estimator, X, Y, cv=kfold)
    print(“Baseline: %.2f (%.2f) MSE” % (results.mean(), results.std()))

    stop = timeit.default_timer()
    print(‘Time: ‘, stop – start)
    Hope I can get your help

  13. slim toueiti April 16, 2020 at 7:36 pm #

    768/768 [==============================] – 0s 32us/step
    accuracy: 73.31%


  14. Sharad Dadhich April 17, 2020 at 6:07 pm #

    Greetings of the day sir
    How to predict Uncertainty in neural networks?

    • Jason Brownlee April 18, 2020 at 5:43 am #

      Neural nets will predict the probability of a class label (uncertainty) by default.

      To predict uncertainty for a numeric value in a regression problem, you can fit multiple final models and use them to get a distribution of expected values – a quick and dirty prediction interval.

  15. Alessandro Fontana May 8, 2020 at 7:29 pm #

    lesson 02 example give me a warning that point here:

    suggestions to avoid it?

    • Jason Brownlee May 9, 2020 at 6:12 am #

      Great work!

      • Alessandro Fontana May 9, 2020 at 4:58 pm #

        “Great work!” …was only a note….. mmmm … 🤔
        … maybe you trained an LSTM to answer some common comments? 😊


  16. Dominique May 15, 2020 at 2:44 am #

    Hi Jason,

    I am running the code examples of Chapter 7 “Evaluate The performance of Deep Learning Models” of your book “Deep Learning With Python.

    I get the following results:
    a) for automatic verification: val_accuracy: 0.7717
    b) for manual verification: val_accuracy: 0.7874
    c) for manual k-fold Cross validation: 70.30% (+/- 6.05%)

    You write in your book that “The gold standard for machine learning model evaluation is k-fold cross-validation”.

    So I don’t understand why it’s with k-fold cross validation that I get the worst result :=(

    Do you have any idea why I get the opposite results to your statement?

    Kind regards,

    • Jason Brownlee May 15, 2020 at 6:07 am #

      It may be a worse result, but it is likely a better estimate of the true performance of the model compared to other methods.

      • dominique May 16, 2020 at 1:28 am #

        Hi Jason,

        Thank you very much for your answer that makes sense now.


  17. Dominique May 16, 2020 at 1:26 am #

    Hi Jason,

    In chapter 8 “Use Keras Models with Scikit-Learn For General Machine Learning” of your book “Deep Learning With Python”.

    When I run the Grid search Neural network, the time of execution is 42 minutes. That’s a bit long as you say in your book it should take 5 minutes.

    I am running the code on a recent iMac with a CPU 3Ghz Intel core i5 6 coeurs with macOS Catalina.

    It’s true that when the program starts I get the message “Your CPU supports instructions that this TensorFlow binary was not compile to use: AX2 FMA.”

    I have installed TensorFlow with anaconda: conda install -c conda-forge tensor flow

    Do you have any suggestion for having a better speed of execution?

    Kind regards,

    • Jason Brownlee May 16, 2020 at 6:16 am #

      One approach might be to run it on AWS EC2 with GPU support.

  18. Dominique May 16, 2020 at 4:28 pm #

    Hi Jason,

    thanks for your answer. I will try.

  19. Dominique May 17, 2020 at 2:41 am #

    Hi Jason,

    In Chapter 11 “Project: Regression Of Boston House Prices” of you book “Deep Learning With Python”, I got the following results:

    I do not see any major differences between Larger and Wider but a clear improvement with the Standardization.

    Thanks for these very didactic examples.

    But I note that the examples with HostonHousing or Iris already take time to execute and I consider them as small set of data. So for biggest dataset obviously the need for more CPU/GPU/TFU hardware is mandatory.

    May I ask you a question?

    I would like to know if it is worthwhile to invest, as an individual, in renting AWS EC2 GPU power to run your examples in the book? isn’t it too costly?

    Kind regards,

    • Jason Brownlee May 17, 2020 at 6:38 am #

      Nice work!

      I recommend start by renting time, it is very cheap when getting started. You don’t need a large/powerful instance to get a lot of improvement.

      Also, many readers use google colab which is free. I don’t know much about it.

  20. Dominique May 17, 2020 at 9:19 pm #

    Hi Jason,

    Thanks for the Colab tip.

    I had a quick try for the wider example.
    On Colab: 1’27”
    On my iMac: 58″

    I will compare later for more complex examples. I have no doubt that Google GPUs are faster than my iMac :=) By the way there is a link on how to upload the local file like the data set for example:

    Kind regards,

  21. Dominique May 20, 2020 at 3:19 pm #

    Hi Jason,

    For lesson 9 “Reduce Overfitting with Dropout regularization” of your Book “Deep Learning with Python”, I got the following results with the code provide in your book:

    Baseline: a) 84,05% (8,23%) b) 85,50% (10,07%)
    Dropout on visible layer a) 87,95% (7,20%) b) 87,00% (6,12%)
    Dropout hidden layers a) 83,60% (8,51%) b) 87,52% (5,25%)

    It seems dropout has the best effect when used only on the visible layer.

    Kind regards,

  22. Dominique May 21, 2020 at 7:09 pm #

    Hi Jason,

    About Chapter 18 “Project Handwritten Digit Recognition” of your Book “Deep Learning with Python”, I got the following results:

    Your book is very cool. We can progress steadily and see concrete results.


  23. Dominique May 22, 2020 at 1:38 am #

    Hi Jason,

    About Chapter 20 “Project Object Recognition in Photographs” of your book “Deep Learning with Python”, I got the following results:

    Better accuracy comes at the expense of execution time :=) I will have to move to Colab or AWS.

    I have a question that I think I am not the only one on this planet that is asking for: I have a full set of photographies on my computer for which I would like to regroup by classification based on face recognition. Do you have any suggestion for a source of information?


    Kind regards,

    • Jason Brownlee May 22, 2020 at 6:10 am #

      Yes, you can use face detection and then train a face recognition system. I show exactly how in this book:

      Also, I think this type of feature is probably built into photo management software these days.

      • Dominique May 22, 2020 at 6:48 pm #

        Hi Jason,


        I had a try on Google Colab for the “Large CNN CIFAR 10” and it’s amazing. The test runs in 6 minutes (after selection of GPU in the notebook parameter) compared to 1 hour on my iMac (which is recent). With an accuracy of 77,80%.

        But with the TFU option Google Colab, the test is too long from the very first epoch.

        Do you have an idea why TFU is of any help?

        Kind regards,

  24. Enrique Aparicio May 22, 2020 at 5:53 am #

    Hi Jason:

    Just tried lesson number 5, and it goes very well.

    Trying to do some of my own I built a database with two coluns
    First column secuencial numbers from 1 to 500, second column 3 times first column
    The answer to this is obvius y= 3 *x

    Modified the program and obtained the following, that seems to me it i not working:

    Loss = 753758,9960
    accuracy = 0.00%

    What is it that I am doing wrong ?

    The modified program is:

    # -*- coding: utf-8 -*-
    Spider Editor
    Created on Thu May 21 12:04:05 2020
    @author: Caro Sertorio

    DATABASE USEDbuilt as follows
    X data is correlative number 500
    Y data is correlative 3 times column x
    a csv table with functio y=f(x) being y=3*x
    np.shape(dataset) is tuple dimension = (499,2)
    from keras.models import Sequential
    from keras.layers import Dense
    import numpy as np
    import pandas as pd

    # Load the dataset
    dataset= pd.read_csv(‘C:/Neural/Data/Tabla x 3.csv’,sep=’;’,header=0)
    dataset = dataset.values
    X = dataset[:,0]
    Y = dataset[:,1]

    # Define and Compile
    model = Sequential()
    model.add(Dense(1, input_dim=1, activation=’relu’))
    # model.add(Dense(10, activation=’relu’))
    model.add(Dense(1, activation=’relu’))
    model.compile(loss=’mean_squared_error’ , optimizer=’adam’, metrics=[‘accuracy’])
    # Fit the model, Y, epochs=10, batch_size=5)
    # Evaluate the model
    scores = model.evaluate(X, Y)
    print(“%s: %.2f%%” % (model.metrics_names[1], scores[1]*100))

    Hope you can comment


    • Jason Brownlee May 22, 2020 at 6:14 am #

      Well done!

      You cannot use relu in the activation function for the output layer. For regression use linear. Also, you cannot use accuracy for regression.

  25. dominique May 24, 2020 at 2:52 am #

    Hi Jason,

    i have just finished your book “Deep Learning With Python”. I found it excellent as the previous one I read “Machine Learning Mastery with R”. I would advise any reader of your blog to buy those books. For me it speeds up my learning curve.

    I published a post on my blog summarizing my experience of your book.

    Thanks very much for all the progress I got up to your books.

    Kind regards,

    • Jason Brownlee May 24, 2020 at 6:14 am #

      Thanks Dominique!

      Writing summaries like you have done is the best way to crystallize what you have learned, great work!

  26. Dominique May 25, 2020 at 3:04 am #

    Hi Jason,

    Today I run your example of code “Text Generation With Alice in Wonderland” you provide in your book “Deep Learning With Python”.

    In fact I replace “Alice in Wonderland” with a famous title from Victor Hugo (a famous French writer) “Les misérables”.

    The total number of characters was 625 332. It took about 6h30 to train the model on my iMAC. But I admit that I was really surprised by the results!

    I published a post to summarize my findings:

    Again thanks for your book,
    Kind regards,

  27. M Muthurajan June 3, 2020 at 6:09 pm #

    lesson 2
    In this lesson, your goal is to install TensorFlow become familiar with the syntax of the symbolic expressions used in TensorFlow programs.

    For example, you can install TensorFlow using pip. There are many different versions of TensorFlow, specialized for each platform. Select the right version for your platform on the TensorFlow installation webpage.

    A small example of a TensorFlow program that you can use as a starting point is listed below:

    When I tried after installing tensorflow, warning message has appeared as follows:
    n error ocurred while starting the kernel
    Warning! ***HDF5 library version mismatched error***
    The HDF5 header files used to compile this application do not match
    the version used by the HDF5 library to which this application is linked.
    Data corruption or segmentation faults may occur if the application continues.
    This can happen when an application was compiled by one version of HDF5 but
    linked with a different version of static or shared HDF5 library.
    You should recompile the application or check your shared library related
    settings such as ‘LD_LIBRARY_PATH’.
    You can, at your own risk, disable this warning by setting the environment
    variable ‘HDF5_DISABLE_VERSION_CHECK’ to a value of ‘1’.
    Setting it to 2 or higher will suppress the warning messages totally.
    Headers are 1.10.4, library is 1.10.5

    General Information:
    HDF5 Version: 1.10.5
    Configured on: 2019󈚧󈚨
    Configured by: Visual Studio 15 2017 Win64
    Host system: Windows󈚮.0.17763
    Uname information: Windows
    Byte sex: little‑endian
    Installation point: C:/Program Files/HDF5

    Compiling Options:
    Build Mode:
    Debugging Symbols:
    Optimization Level:

    Linking Options:
    Statically Linked Executables: OFF
    LDFLAGS: /machine:x64
    Extra libraries:

    C: yes
    C Compiler: C:/Program Files (x86)/Microsoft Visual Studio/2017/Community/VC/Tools/MSVC/14.16.27023/bin/Hostx86/x64/cl.exe 19.16.27027.1
    Shared C Library: YES
    Static C Library: YES

    Fortran: OFF
    Fortran Compiler:
    Fortran Flags:
    H5 Fortran Flags:
    AM Fortran Flags:
    Shared Fortran Library: YES
    Static Fortran Library: YES

    C++: ON
    C++ Compiler: C:/Program Files (x86)/Microsoft Visual Studio/2017/Community/VC/Tools/MSVC/14.16.27023/bin/Hostx86/x64/cl.exe 19.16.27027.1
    C++ Flags: /DWIN32 /D_WINDOWS /W3 /GR /EHsc
    H5 C++ Flags:
    AM C++ Flags:
    Shared C++ Library: YES
    Static C++ Library: YES

    JAVA Compiler:

    Parallel HDF5: OFF
    Parallel Filtered Dataset Writes:
    Large Parallel I/O:
    High‑level library: ON
    Threadsafety: OFF
    Default API mapping: v110
    With deprecated public symbols: ON
    I/O filters (external): DEFLATE DECODE ENCODE
    Direct VFD:
    Packages w/ extra debug output:
    API Tracing: OFF
    Using memory checker: OFF
    Memory allocation sanity checks: OFF
    Function Stack Tracing: OFF
    Strict File Format Checks: OFF
    Optimization Instrumentation:

    Can you please help me.

  28. BANALA SARITHA June 19, 2020 at 3:34 pm #

    Lesson 5: am getting error numpy is not defined

    Can anyone help me

  29. Ashish Soni September 11, 2020 at 9:15 am #

    Lesson 01: Introduction to Theano

    the output for the result variable is 4.0

  30. Ashish Soni September 11, 2020 at 9:25 am #

    Lesson 02: Introduction to TensorFlow

    output for c : 4.0

  31. Abdulla Mamun October 19, 2020 at 9:18 pm #

    Hi Jason,
    its my first day to start with theano.
    I have run the code in colab and found the result of two variables sum of a and b is 4 .
    kind regards

  32. Francisco Gonçalves November 9, 2020 at 8:02 am #

    Hi Jason,
    I got the error below in Lesson 06. Could you help me, please?

    NameError Traceback (most recent call last)
    11 # create classifier for use in scikit-learn
    —> 12 model = KerasClassifier(build_fn=create_model, epochs=150, batch_size=10)
    13 # evaluate model using 10-fold cross validation in scikit-learn
    14 kfold = StratifieldKFolder(n_splits=10, shuffle=True, randon_state=seed)

    NameError: name ‘KerasClassifier’ is not defined

    • Jason Brownlee November 9, 2020 at 1:13 pm #

      The error suggest that Keras is not installed or that you have not imported the class (e.g. skipped some lines).

      • Francisco Gonçalves November 16, 2020 at 10:39 am #


        • Francisco Gonçalves November 23, 2020 at 8:20 am #

          Hi Jason,
          Here you can see the output of lesson 6:


          The result appear with some delay. Is it normal?


          • Jason Brownlee November 23, 2020 at 10:50 am #

            Nice work.

            Yes. It may take time to complete the model evaluation.

  33. Greg Pautsch November 11, 2020 at 3:09 am #

    Installing Theano

    I typed pip installed Theano into the Anaconda powershell and it seemed to work

    Then in Anaconda, I typed

    import theano

    from theano import tensor

    and received the following warning

    WARNING (theano.configdefaults): g++ not available, if using conda: conda install m2w64-toolchain

    C:\Program Files\Anaconda3\lib\site-packages\theano\ UserWarning: DeprecationWarning: there is no c++ compiler.This is deprecated and with Theano 0.11 a c++ compiler will be mandatory

    warnings.warn(“DeprecationWarning: there is no c++ compiler.”

    WARNING (theano.configdefaults): g++ not detected ! Theano will be unable to execute optimized C-implementations (for both CPU and GPU) and will default to Python implementations. Performance will be severely degraded. To remove this warning, set Theano flags cxx to an empty string.

    “dot” with args [‘-Tps’, ‘C:\\Users\\gpautsch\\AppData\\Local\\Temp\\tmpuavj5jyx’] returned code: 1

    stdout, stderr:



    WARNING (theano.tensor.blas): Using NumPy C-API based implementation for BLAS functions.

    I then typed in the rest of your suggested code and it worked

    # declare two symbolic floating-point scalars

    a = tensor.dscalar()

    b = tensor.dscalar()

    # create a simple expression

    c = a + b

    # convert the expression into a callable object that takes (a,b)

    # values as input and computes a value for c

    f = theano.function([a,b], c)

    # bind 1.5 to ‘a’, 2.5 to ‘b’, and evaluate ‘c’

    result = f(1.5, 2.5)


    The result was 4.0

    Should I be doing anything with the warning?

    • Jason Brownlee November 11, 2020 at 6:52 am #

      Perhaps ignore the warnings for now.

      A lot of these new libs dump a lot of stuff to console – I think it is a terrible practice.

  34. Francisco Gonçalves November 23, 2020 at 8:19 am #

    Hi Jason,
    Here you can see the output of lesson 6:


    The result appear with some delay. Is it normal?


  35. Azerul Azlan December 8, 2020 at 12:44 am #

    Hi Jason,

    Lesson2:Introduction to TensorFlow

    Answer is 4.0

    For lesson 1 I met the error that wrote:
    Traceback (most recent call last):
    File “”, line 1, in
    import theano
    File “C:\Users\Azerul Azlan\pillow\”, line 2, in
    from theano import tensor
    ImportError: cannot import name ‘tensor’ from ‘theano’

    How to fix it?

    • Jason Brownlee December 8, 2020 at 7:44 am #

      Perhaps you need to update your version of theano?

      • Azerul Azlan December 9, 2020 at 3:03 pm #

        How to update theano?

        • Jason Brownlee December 10, 2020 at 6:18 am #

          As follows:

  36. Nicholas December 18, 2020 at 8:44 pm #

    Numpy Error in Lesson 05: Develop Your First Neural Network in Keras

    below lines will solve it

    import numpy as np
    # Load the dataset
    dataset = np.loadtxt(“pima-indians-diabetes.csv”, delimiter=”,”)

    • Jason Brownlee December 19, 2020 at 6:16 am #

      Sorry to hear you got an error.

      Thanks for sharing.

Leave a Reply