in Blog

June 18, 2020

Machine Learning Frameworks


Edwin Lisowski

CSO & Co-Founder

Reading time:

10 minutes

Last time, we talked about machine learning software. Now it’s time for a similar subject. We want to show you some of the most popular machine learning frameworks. Although framework and software are strictly related terms, they don’t mean the same thing. What’s the difference between machine learning software and machine learning framework? And what are the most popular machine learning frameworks? It’s time to find out!

Before we begin, let’s clarify the first crucial question. What’s the difference between software and framework? Software is simply a program, an application, or a set of instructions that run on a machine, typically on a computer or a mobile device. A framework is also a software, but it was written in order to use it to develop other software. In other words, it’s software designed to build software.

Usually, frameworks are open source and the same rule applies to the machine learning world, where almost all the machine learning frameworks are 100% free to use, even for commercial purposes.

How are Machine Learning frameworks used?

Actually, frameworks have significantly improved the pace at which machine learning grows. They allowed many companies and people to develop their own programs and algorithms, without the need to invest thousands of dollars to build necessary infrastructure from scratch.

New machine learning frameworks are mushrooming today. That’s due to increased awareness of the technology and its possible applications. The more people use ML frameworks, the more applications they come up with, and, as a result, new solutions are needed.

For instance, in 2017, Amazon and Microsoft have announced Gluon[1], a new open-source machine & deep learning interface. According to Amazon, it’s an API designed for defining machine learning models using a collection of pre-built, optimized neural network components.

What is a Machine Learning framework?

We can say that the machine learning framework is an interface, library, or tool designed for the developers in order to build machine learning models and algorithms more quickly and easily. It’s a massive simplification, but it’s a bit like the LEGO bricks. You can build your machine learning models using a collection of pre-built, optimized components (in our example, bricks) to build your projects (buildings). ML frameworks help developers build their models in a precise, transparent, and concise way.

Machine learning frameworks reduce the complexity of coding and training ML models and algorithms, thus making this field more accessible for less experienced developers.

Now, what should you keep in mind when choosing the best framework for yourself? There are a couple of aspects you should take into account:

  • Good ML framework is performance-oriented and helps you in building your models
  • It’s easy to learn and understand
  • It should have extensive documentation and community that will help you with any difficulties you may face
  • It should be an open-source platform that can be used for commercial purposes
  • Your framework should provide you with massive computing power and allow you to use large amounts of data (that’s especially important when you want to train deep learning models, which are much more complicated than typical ML programs)

If you read our last article about machine learning software, you already know some of the most popular machine learning frameworks. In that article, we mentioned:

  • Tensor Flow

All of these platforms are, in essence, machine learning frameworks. But there are far more of them on the market! Today, we will show you another ten ML frameworks you can freely use for your machine learning projects. Here we go!

Machine Learning frameworks examples

1. Theano

Theano is often perceived as a major Tensor Flow competitor. In fact, both of them are extensively used by the researchers in the machine learning consulting services and deep learning domain. Theano is a Python-based library that allows users’ to define, optimize, and evaluate mathematical expressions evolving multi-dimensional arrays.

Theano can be deployed on a single GPU and allows transparent use of GPU to perform data computations. This solution has many significant advantages. For instance, many open-source deep learning libraries, such as Keras, are built on top of Theano. It has high-level wrappers such as aforementioned Keras and Lasagne, which increases its usability. It can be, however, troublesome on AWS. And when you’re building large ML models, you have to take long compile times into account.

2. Caffe

It’s another deep learning framework, originally developed at the University of California. According to its website, Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by Berkeley AI Research (BAIR) and by community contributors. Yangqing Jia created the project during his Ph.D. at UC Berkeley. Caffe is released under the BSD 2-Clause license[2].

The Caffe framework supports many different types of deep learning architectures geared towards image classification and segmentation. It supports CNN, RCNN, LSTM, and fully connected neural network designs. Caffe is being commonly used in academic research, start-up prototypes, and even large-scale industrial applications in vision, speech, and multimedia[3].

When it comes to Caffe, ML models and optimization are defined by configuration without hard-coding. It’s also quite speedy software! Caffe can process over 60M images per day with a single NVIDIA K40 GPU. That’s 1 ms/image for inference and 4 ms/image for learning! Last but not least, Caffe comes with extensive documentation and a supportive community that constantly contributes to the development of this software. What’s interesting, Google’s DeepDream depends on Caffe Framework.

3. H2O

H2O is an open-source machine learning platform with linear scalability. It is an artificial intelligence tool that is business-oriented and helps in making decisions based on data and enables the user to draw insights. H2O has an industry-leading AutoML functionality that automatically runs through all the algorithms and their hyperparameters to produce a leaderboard of the best models. It is mostly used for predictive modeling, risk and fraud analysis, insurance analytics, advertising technology, healthcare, and customer intelligence.

H2O can be used for the Python and R machine learning projects.

4. Torch

Torch is an open-source machine learning library, but also a framework, and a script language based on the Lua programming language. It provides a wide range of algorithms, designed especially for deep learning. It has some of the interesting features:

  • A powerful N-dimensional array
  • Useful interface to C, via LuaJIT
  • Linear algebra routines
  • A neural network, and energy-based model
  • Fast and efficient GPU support
  • Embeddable, with ports to iOS and Android back-ends

Currently, Torch is no longer in active development[4]. PyTorch is a worth-mentioning alternative. With PyTorch, there is no need to write code in the Lua programming language. Read our previous blog post about machine learning software to find out more about PyTorch.

5. Azure ML Studio

Its full name is Microsoft Azure Machine Learning Studio (classic). It’s a drag-and-drop collaboration tool that is used to compile, test, and implement predictive data analytics. Azure publishes models as web services that can be easily used by custom applications or business intelligence tools such as Excel[5]. This Framework allows Microsoft Azure users to create and train models, then turn them into APIs that can be consumed by other services.

Azure ML Studio can be used for anomaly detection, classification, clustering, regression, statistical functions, text analytics, and computer vision.

6. Microsoft Cognitive Toolkit

It’s an open-source toolkit made for commercial deep learning projects. It describes neural networks as a series of computational steps via a directed graph. Although Microsoft has built it with deep learning in mind, it can be used to process huge amounts of data for other machine learning models. It’s frequently recommended for specialists who want to take their first steps in the world of deep learning, especially because it’s highly customizable and supports multi-machine back ends.

7. Firebase ML Kit

Google introduced this Firebase ML kit to help developers integrate machine learning models into mobile apps without a lot of technical expertise and coding. According to the Firebase website: It brings Google’s machine learning expertise to Android and iOS apps in a powerful yet easy-to-use package. Whether you’re new or experienced in machine learning, you can implement the functionality you need in just a few lines of code. There’s no need to have deep knowledge of neural networks or model optimization to get started. On the other hand, if you are an experienced ML developer, ML Kit provides convenient APIs that help you use your custom TensorFlow Lite models in your mobile apps [6].

If you want to start working on mobile machine learning projects, it’s one of your best bets.

8. Apache Spark MLlib

MLlib is Apache Spark’s scalable machine learning library. It consists of common machine learning algorithms and utilities, i.a.: classification, regression, clustering, collaborative filtering, and dimensionality reduction[7].

According to their website, ML workflow utilities include:

  • Feature transformations
  • ML Pipeline construction
  • Model evaluation and hyper-parameter tuning
  • ML persistence: saving and loading models and Pipelines

9. Apache Singa

It is an Apache project designed for developing an open-source machine learning library. It provides a flexible architecture for scalable distributed training, and it is extensible to run over a wide range of hardware. It’s primarily used for training big deep learning models over large datasets. Apache Singa supports Convolutional Neural Networks (CNN), Boltzmann Machine (RBM), and Recurrent Neural Networks (RNN).

10. MOA (Massive Online Analysis)

MOA [8] is the most popular open-source framework designed for data stream mining. It includes a collection of machine learning algorithms (classification, regression, clustering, outlier detection, concept drift detection, and recommendation systems) and various tools for evaluation. This framework is written in Java. What’s especially important, MOA can be extended with new mining algorithms, and new stream generators or evaluation measures.

Machine Learning frameworks – conclusion

When it comes to machine learning frameworks, you have plenty to choose from, whether you’re after machine learning, deep learning, data engineering services, or data mining. But what to do, when you want to start your machine learning project, but you lack the necessary knowledge or experience? Outsourcing is by far your best bet. You don’t have to grasp all that vast knowledge to get the benefit of it.

With Addepto Team by your side, any machine learning project, even the most complex one, is at hand! Drop us a line or call us, and let’s start working on your ML endeavors.

Also check out our machine learning consulting services to learn more.


[1] Dr. Matt Wood. Introducing Gluon: a new library for machine learning from AWS and Microsoft. Oct 12, 2017. URL: Accessed Jun 18, 2020.

[2] Berkeleyvision. Caffe. URL: Accessed Jun 18, 2020.

[3] Wikipedia. Caffe (software). URL: Accessed Jun 18, 2020.

[4] Wikipedia. Torch (machine learning). URL: Accessed Jun 18, 2020.

[5] Microsoft. ML Studio (classic) vs Azure Machine Learning studio. URL: Accessed Jun 18, 2020.

[6] Google. ML Kit for Firebase. URL: Accessed Jun 18, 2020.

[7] Apache. Apache Spark MLlib. URL: Accessed Jun 18, 2020.

[8] Waikato. Machine Learning for Data Streams. URL: Accessed Jun 18, 2020.



Machine Learning