Cloud Computing for Data Analysis

108 Chapter 4: Challenges and Opportunities in Distributed Computing

GPU Workflow

A key concept in the GPU program is the following steps to code to the GPU from scrat.

1. Create a Vectorized Function

2. Move calculations to GPU memory

3. Calculate on the GPU

4. Move the Calculations ba to the host

Here is a contrived example below that is available in this Google Colab Notebook²⁴⁷.

1 from numba import (cuda, vectorize)

2 import pandas as pd

3 import numpy as np

4 from sklearn.preprocessing import MinMaxScaler

5 from sklearn.cluster import KMeans

6

7 from functools import wraps

8 from time import time

9

10 def real_estate_df():

11 """30 Years of Housing Prices""" ...