1 / 7

Data Analysis in Class Projects: How to Process Data, Split Sets, and Avoid Overfitting

In this Kaggle class project setup guide, learn how to handle your data effectively with a literature review, basic statistics, and feature analysis. Discover how to split your data into training and test sets, ensuring proper labeling and class representation. Avoid overfitting by building your system only using the training set and analyzing the error versus complexity trade-off. Explore tasks like literature reviews, data statistics, data splitting strategies, and the optional development and cross-validation sets. Enhance your understanding of dog vs. cat classification example with image and feature data.

ghalib
Download Presentation

Data Analysis in Class Projects: How to Process Data, Split Sets, and Avoid Overfitting

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Mebi 591D –BHI Kaggle Class Project Setup http://winter2014-mebi591d-kaggleclass.weebly.com/

  2. Project Setup (I.) • You have your data, now what? • Literature Review • Basic statistics of data • How many training/test instances • How many categories? • Frequency of features • How many non-zero features, etc….. Due next week: 5-10 min, present your problem + lit + data

  3. Examples Dogs vs. cats (features given) • 1000 dogs, 278 cats • 10000 features • 1356 non-zero features • Dog categorization (images given) • 7 classes (#s of each) • #black and white images • # of images with {1 dog, 2 dogs, 3 dogs, etc}

  4. Project Setup (II.) 3. Split Data - training set vs. test set - if you don’t have test answers > label yourself > randomly split out your own test set - make sure to sample each class 4. Build system using training set only! - Why shouldn’t I evaluate on my test set?

  5. evaluating on same set as testing will lead to 100% accuracy – but you are overtraining!! True function (due to unseen examples and noise) will look differently! Overfitting • … Another reason why you should not evaluate on your data you’ve trained on Fit to a polynomial: Y = a0 + a1*x + a2*x2 + … + a20*x20 Evaluate on ALL Data What would happen?

  6. Error vs. Complexity true error Error training error Complexity

  7. Tasks • Literature review (next week) • Data statistics (next week) • Split data • Training vs. split • (optional) development set • (optional) cross-validation sets

More Related