Nuts & Bolts of Machine Learning

For more projects visit Github

Useful Deep Learning Libraries

Image Credit: https://www.analyticsindiamag.com

If you are a new to the field of machine learning, you can get overwhelmed with the variety of sub-fields available under the umbrella of machine learning. For each subfield, there are specific libraries to help you with getting the task done. In this blog, we will glance over the different libraries and their use cases.

General Purpose Machine Learning Libraries

Scikit-Learn

Scitkit-learn provides all popular machine learning algorithms and functions for data transformation to error analysis pre-built into it. It is a recommended library for getting started with machine learning. One can tweak the parameters in algortihms and see the changes in the performance of model.

Tensorflow

Most of the people take tensorflow as deep learning library but at its core it just converts mathematical operations into a computational grpahs. Hence, any machine learning algorithm can be developed in tensorlfow. The implementations of Logistic Regression and Linear Regression in tensorlfow can be easily found in web.

Working with Videos/Images

OpenCV

OpenCV is the most popular library for working with images and videos. It provides all basic operations to be executed on images. It enables us to control webcam and take images or video-feed from it. Hence, it becomes suitable for live projects of object detection, face recognition etc.

Scikit-Image

It provides optimized image transformation operations.

DLib

It provides deep learning based implementations for face detection and object detection.

Working with Text Data

NLTK

NLTK (Natural Language TooKit) provides basic operations for text pre-processing.

Spacy

Spacy provides all text pre-processing functions along with deep learning based models for various NLP tasks.

Stanford-NLP

It is developed in Stanford as evidnet from name. It is a cutting edge library which features latest developments in the field of NLP.

Gensim

Gensim contains various trained models for word-embeddings and topic-modelling.

Reinforcement Learning

OpenAI Gym

For most of reinforcement leraning tasks, an enviornment is required where agent can perform actions and learn. Gym provides various environments where one can train agents. The environment includes atari games, robotic simulations and alot of other fancy things.

PyGame

If you want to create a new environment (or a new game) and train agent in it, you can build it with PyGame.

Keras-RL

Keras-RL is deep reinforcement learning library. It provides deep architectures which can be trained in Gym or other environment. It features popular algorithms such as DQN.

Deep Learning Libraries

Theano

Theano was developed by MILA and it is predecessor of all deep learning frameworks. But now MILA has stopped its development after 1.0 release.

Tensorflow

Tensorflow is the most popular library for machine learning. It emphasized on the concept of computational graphs. It is well suited for distributed computing. It is supported by Google.

Pytorch

Pytorch is being developed by Facebook and it is the main rival of Tensorflow. Pytorch also provides a dynamic graph execution.

Keras

Keras is a high-level wrapper on existing deep learning frameworks. It supports theano, tensorflow and mxnet as its backends. It is recommended for beginners as deep learning architectures can be made very easily, quickly and without in-depth knowledge of neural networks.

Other Deep Learning Libraries:-

H2O

Supported by H2O.ai which provide several deep learning solutions such as Driverless AI

Caffe

It was originally developed by Berkley AI Research (BAIR) group, but now it is advanced by Facebook

CNTK

A deep learning library provided by Microsift

MXNet

MXNet is supported by Amazon

Most of the deep learing libraries have two different versions for CPU and GPU. The GPU version of the library takes advantage of CUDA acceleration and performance gain is enormous. So, if you have Nvidia graphics card, setup CUDA and install gpu version of the library.

Other Supporting Libraries

Numpy

Numpy is a library for processing numerical data. It is useful in data pre-processing step. It gives good performance when vectorization is used. It also provides very optimized implementations of common mathematical operations.

Pandas

Pandas is used for mix data as it can handle strings and numerical data as well. Several useful transformations are already implemented (such as fixing null values, one-hot encoding) and easily handles datetime data. It has become popular for data preprocessing.

Scipy

Scipy provides complex mathematical operations such as convolution and matrix inversions along with various statistical functions. It also serves as backend of Scikit-Learn and Scikit-Image.

Matplotlib

Matplotlib is most popular library for data visualization. Data visualization is helpful in analyzing the data during EDA (Exploratory Data Analysis).

Other popular libraries for data visualization:-

Bokeh, Plotly, Seaborn, Graphviz