AI Application Boost with RAPIDS GPU Acceleration

Explore cuDF, a RAPIDS library that migrates pandas workflows from CPU to GPU with a pandas-like API for data manipulation and analysis.

Install rapids on Google Colab via a single command or a repository script, selecting GPU, CUDA, and Python versions. Save a copy of the notebook to Google Drive when customizing.

Explore pandas on the cpu and rapids cuDF on the gpu by building a data frame, adding columns, and summing a column, revealing nearly identical code across platforms.

learn how to create cudf series and data frames, convert pandas dataframes, and perform head, sort_values, and iloc indexing on the gpu.

Explore basic commands for data filtering with conditions and booleans, including df.query and missing data handling, then compare numpy and cupy for CPU and GPU arrays, and review describe statistics.

Apply user defined functions to data frames with CUDA, using df.apply, apply rows, and apply chunks for GPU acceleration and missing-value handling. Create UDFs like add for row operations.

Explore the performance comparison between rapids on GPU and pandas on CPU, testing value counts, concatenation, group by, merge, and string operations on datasets with 10 million rows.

Implement linear regression with cuML using CUDA and CuPy, training with fit and predicting with predict on CuPy data. Compare results to sklearn and visualize with matplotlib.

Explore ridge regression on the diabetes dataset using real data, comparing sklearn and cuml implementations with gpu acceleration, including train-test split and score evaluation.

Apply hyperparameter tuning with grid search cv in sklearn to find the best alpha for the ridge algorithm, improving the r2 score from 0.37 to 0.49 on a small dataset.

Compare rapids with sklearn using qml benchmark runners and the speedup comparison runner to automate performance tests across varied data sets, repetitions, and CPU and GPU settings.

Compare gpu and cpu performance for linear regression and clustering with rapids acceleration, showing speedups across data sets and highlighting practical results from benchmark tests.

Create and visualize benchmark speedups using a data frame and pivot tables to compare RAPIDS against scikit learn across linear regression, kmeans, and random forest on multiple datasets.

Set up NVIDIA RAPIDS on Google Colab by installing libraries, selecting a GPU, and verifying cudf 23.12.01 and cupy 13.0.0 versions for data analysis readiness.

Load a census dataset using CDF for GPU-accelerated processing, inspect data with describe and missing values, visualize income distribution, and split into features x and target y for modeling.

Apply one hot encoding to categorical features with a column transformer, turning banana and apple into binary columns. See the dataset grow to 108 columns with QML pre-processing.

Standardize features with mean and standard deviation using a standard scalar, then encode categorical values, split data into train and test (15% test, 42 random state), and save preprocessed data.

Load the census dataset, split into train and test, and train logistic regression and kNN using rapids acceleration on the gpu, then assess accuracy and confusion matrix.

Apply random forest and SVM using NVIDIA RAPIDS with QML on a real dataset, train the models, generate predictions, and evaluate with accuracy and a confusion matrix.

Explore how to run a gpu-accelerated workflow in Colab using NVIDIA RAPIDS to download, preprocess, balance, and encode a large US accident dataset for severity prediction.

Apply one hot encoding to predictive attributes, drop the class, standardize, then split 85/15 and run a k neighbors classifier in rapids, achieving 62% accuracy.

Discover how dask and rapids enable distributed processing in Python, forming clusters of clients and workers to scale data frames and arrays across CPUs and GPUs.

Create a local cluster for accelerated data processing by installing Rapids, using Dask-CUDA, and configuring a local CUDA cluster in Google Colab to run Dask with a single GPU.

Distribute a 100,000 x 100 matrix across GPUs or CPUs using Dask array, with CuPy random state and chunking, then perform SVD and persist results to GPU memory.

Learn to integrate Dask and cuDF to partition data across GPUs and CPUs, compute across partitions, and export results to CSV.

Explore distributed machine learning with Dask and cuML, creating a local CUDA cluster, distributing data across workers, and training a random forest on CPU and GPU.

Explore parallel training of random forests with dask and cuml alongside scikit-learn, comparing cpu and dask workflows, with asynchronous compute and multi-gpu acceleration for fast predictions.

Review the basics of gpu, cuda fundamentals, and the benefits of gpu acceleration over cpu. Explore rapids api with cudf, cuml, and dask for multi-gpu or multi-core performance.

Discover artificial intelligence and data science concepts in a bonus session of the ai application boost course, covering machine learning, deep learning, computer vision, natural language processing, and algorithms.