Intro

This blog is about using Tensorflow with Python for machine learning. I'll start with giving some background information in order for the reader to better relate to what's coming up. I'll try to keep things short and to the point.

Academic background & work history

I have Ph.D. in computational materials physics and in the course of my studies, did numerical simulations about fracture of disordered media (paper specifically). The purpose of the research was to gain insight into fracture of paper so that paper machines could be run at higher speeds.

I started my simulations with Cray X-MP and Convex computers which had vector processors inside. Programming was easy since Cray's compiler output vectorized code, provided that usual design criteria for vectorization were met with.

In the middle of the graduate studies, computation platform changed to Cray T3E, which consisted of separate computer systems connected with fast optical fibre communications. For programming, libraries like PVM and MPI were used for distributing computations across CPUs.

Since 1998, I've been employed by Nokia, and involved in research, R&D, business management, and consultation roles in the area of telecommunications. Most recently, I have been involved in research related to 5G management and orchestration (M&O). This is an area which involves Machine Learning (ML) and aspects of Artificial Intelligence (AI), which have made my earlier experience with numerical programming relevant again. The big difference is that now one can do impressive computations with a general-purpose CPU and GPU in a mass-market computer system.

Programming environments

I started scientific programming with plain C. At Nokia, I used C++ for years, mostly for information transformations. During a short sabbatical, I read a book about Common Lisp and was impressed enough to switch to CL as my development platform. Later on, I switched to Clojure due to JVM integration and for a vibrant ecosystem with a great selection of libraries and a great package management system. I was interested in Haskell, and liked a lot some things like monads and comprehensions. I was put off by appalling package management system and a perception of clunky syntax compared to Lisp style syntax.

I was not interested in Python until had to give a hand to trainees with programming. I thought that slowness of Python's native data types (100+ times slower than native C) would be a showstopper. In getting to know Python, I got acquainted with Numpy library for efficient computations which mostly removed the main obstacle in terms of performance. I liked Haskell-style link comprehensions as well as ease of using functional programming concepts like map, reduce and lambda functions. Overall, the functional programming aspects of Python made it easy to switch from Clojure.

I switched to Python as my main programming environment and noticed a clear improvement in productivity for analytics and ML. The next frontier was using artificial neural networks (ANNs) in machine learning, which led naturally to starting to use Tensorflow.

Comments

Post a Comment

Popular posts from this blog

Business of Machine Learning

Image
The role of Machine Learning (ML) in business is an interesting topic. ML provides a powerful tool set for understanding data, but that is not enough. A fundamental tenet of strategy is differentiation, or finding a sustainable competitive advantage. We shall try to clarify basic alternatives below. Suunto dive computers use mathematical model for nitrogen saturation of human body. For deep dives or challenging conditions, using multiple computers provides safety. Exclusive access to data This is the most talked about approach at the moment: a company has exclusive access to data and performs value adding operations in them. In the simplest case, the data in question are related to normal operations of the company, and may be Customer Relationship Management (CRM) data or enterprise business data, for example. Machine learning can be used to automate business processes using data. A second type of exclusive data relates to a service product the company is offering. It may b
Read more

Ethics and machine learning

Machine learning has many new aspects that need to be understood by the society since it is expected to be widely used in the society. A variant of machine learning called supervised learning is used for applications like object detection in pictures as well as for classification. Let's take a quick look at some of the basic concepts below. Model Model defines how input data is processed. Depending on the use case, input data may consist of digital pictures, digitized sounds, or numerical data, for example. The model consists of mathematical operations which transform the input data into desired output (e.g. classification "cat" / "dog" in picture). There are many algorithms which can be used for the model, ranging from simple mathematical formulae to fashionable deep learning networks. A model is basically neutral and generic, so that an individual model may be used for classifying cats vs. dogs on the one hand, or types of trees on the other. For dema
Read more

On y va

Getting started with TensorFlow is relatively easy, assuming one is familiar with ANNs. TensorFlow's website  provides good examples to get one started. For an example on how ANNs are implemented with TensorFlow, go straight for "Get started" => "Deep MNIST for experts". Installation of TensorFlow on a computer without GPU support is easy, following the instructions on the website. Server installation in virtualenv, Anaconda installation using own package management system. One should note that  server installation is not visible in Anaconda if run on same system, one needs separate installs in this case. Below, I'll describe my experiences with GPU (GeForce GTX 1070 on Ubuntu 16.04 LTS) as well as notes about running TensorFlow based programs with both GPU and CPU. GPU installation Installation of TensorFlow with GPU support is a bit trickier, primarily because TensorFlow relies on NVIDIA CUDA  library. This in itself is not a problem, the challe
Read more