Accelerate Machine Learning Workloads: K-means and GPairs Algorithms

Discover efficient methods for implementing the Pairwise and Black-Scholes algorithms using tools included in the AI Tools, powered by oneAPI.

The Pairwise distance applications take a set of multidimensional points and compute the Euclidean distance, cosine distance, or both, between every pair of points.

The Black-Scholes algorithm computes the price of a portfolio of options using partial differential equations. Parallel computations calculate each option price independently of the others.

Sample code on Intel® Developer Cloud (you need an account to participate) shows how these algorithms are implemented using the AI Tools for calculations.

Topics covered during the workshop include:

• A walk-through of the code that implements the Pairwise and Black-Scholes algorithms
• An introduction to Intel® Extension for Scikit-learn*
• An examination of the cosine distance algorithm using Pairwise and Intel Extension for Scikit-learn
• Visualization of the Pairwise and Black-Scholes algorithms using matplotlib
• Compilation and running of the same algorithm code samples on CPU and GPU offload

0:00 Introductions

1:56 Agenda

5:03 Programming challenges

6:05 Introducing oneAPI

7:33 AI Tools

9:10 Intel® VTune™ Profiler

10:28 Intel® Advisor

12:57 Find effective optimization strategies

13:53 Learn more about Intel Developer Cloud for oneAPI

16:19 Get started

22:37 Data parallel essentials for Python*

26:25 Data parallel control

28:33 Compute follows data

31:30 Programming model

32:35 Numba-dpex

36:00 Example of automatic offload using an @njit decorator

37:52 Example of an explicit parallel for loop using an @njit decorator

38:45 Example of an @dppy.kernel decorator

41:34 What categories of AI are covered?

42:13 Types of machine learning

43:35 Types of supervised learning

44:29 Types of unsupervised learning

45:28 Classification and regression

46:36 Supervised learning overview

47:24 Regression: Numeric answers

47:59 Classification: categorical answers

49:08 What is classification?

50:08 What is needed for classification and an example

55:30 Manhattan distance example

56:05 Introduction to patching

57:02 Patching examples

58:08 Pairwise distance example

1:00:12 Cosine distance with scikit-learn

1:02:00 Correlation distance with scikit-learn

1:03:40 Module 1: Introduction to numba-dpex

1:13:38 Module 2: Introduction to Data Parallel Control (dpctl)

1:27:01 Module 3: Pairwise distance algorithm using numba-dpex

1:44:00 A brief overview of the Black-Scholes algorithm using numba-dpex

1:55:38 Q&A