Jump in and drown in all the data
Here is a list of data avaiable.
“Data is the crude oil of the 21st Century, and analytics is the combustion engine.” –Gartner
I’d like to see how different people work on the same set of data:
https://www.ted.com/talks/joseph_redmon_how_a_computer_learns_to_recognize_objects_instantly#t-286801”> VIDEO</a>: Joseph Redmon’s YOLO (You Only Look Once) algorithm recognizes over 80 categories of objects real-time in videos. On a laptop. Based on the University of Washington’s open-source Darknet system.
MNIST Number Images
Instead of downloading yourself, note that the Floydhub.com has these image datasets already on their servers for Machine Learning code use:
On the website of the “Godfather of ML”, Yann Lecun)</a> is the “hello world” of deep learning – 55,000 28x28 pixel images of hand-written numbers (from 0 thru 9). Each image is labeled with the number written in the image. The “NIST” in “MNIST” is for the US National Institute of Technology.
this lists methods by their error rate.
MNIST using a “flashlight” visualization by Tensorboard by Dandelion at the TensorFlow Dev Summit Feb. 2017.
The MNIST dataset comes pre-loaded in Keras, in the form of a set of four Numpy arrays, loaded using this code that references two sets of data – the training set and testing set.
from keras.datasets import mnist (train_images, train_labels), (test_images, test_labels) = mnist.load_data()
The “shape” of an array is the number of items and pixel height and width:
train_images.shape (60000, 28, 28)
COCO is a new image recognition, segmentation, and captioning dataset. It has 300,000 images containing multiple objects per image. 80,000 object categories.
Imagenet VGG Very Deep 19 19 weight layers pre-trained Convnet model
CALTECH 101/256 contains pictures of objects belonging to 101/256 categories
CIFAR 10/100 Subset of 80 million tiny images dataset (cats, horses, airplanes, etc.)
Cats vs Dogs Redux: Kernels Edition Dataset for Kaggle’s famous Dogs vs Cats competition
KONECT (the Koblenz Network Collection) from the Institute of Web Science and Technologies at the University of Koblenz–Landau collects large network datasets of all types in order to perform research in network science and related fields.
Google digitized (scanned) all the books in the 20th century and turned them into n-grams at
https://books.google.com/ngrams/ with counts how often each word occurred in all books.
Wordnet defined affect scores – a mood score.
http://www.makeovermonday.co.uk/data has one (of 52) visualization makeover every week.
IEX (Investors Exchange) has real-time stock exchange.
Azure - Community content are in the Cortana Gallery.
Google Big Data
us_budget has dollar outlays of each bureau within all agency (branch) of the US government, by year from 1962 to 2021
Allen Institute (ai2)
http://news.google.com/archivesearch has 200 years of archives
http://www.ibiblio.org/slanews/internet/intarchives.htm has links to global archives
http://searches.rootsweb.ancestry.com/ssdi.html Roots web
http://search.ancestry.com/search/db.aspx?dbid=3693 US Social Security Death Masterfile Index goes from 1935-2014
http://www.worldcat.org/default.jsp “lets you search the collections of libraries in your community and thousands more around the world.”
Zip codes by state, latitude, longitude
First names registered in each state, by year, in the US from Google Big Data
Musicbase from a game
- Reduction (generalize synonyms)