Published
Edited
Jun 3, 2021
1 star
cars = (await require('vega-datasets@1'))['cars.json']()
keys = ['Cylinders', 'Displacement', 'Horsepower', 'Weight_in_lbs', 'Acceleration', 'Miles_per_Gallon', 'Name', 'Origin']
cars_data = {
var cars_data = cars_filtered.map((row) =>
[[row.Cylinders, row.Displacement, row.Horsepower, row.Weight_in_lbs, row.Acceleration, row.Miles_per_Gallon],
row.Origin, row.Name]);
return cars_data;
}
data_split = {
var cars_shuffled = shuffleArray(cars_data);
const train_count = 0.8*cars_shuffled.length; // we will be using 80% of the data for training
var X_train = cars_shuffled.slice(0, train_count).map(x => x[0]);
var X_test = cars_shuffled.slice(train_count).map(x => x[0]);
var Y_train = cars_shuffled.slice(0, train_count).map(x => x[1]);
var Y_test = cars_shuffled.slice(train_count).map(x => x[1]);
var names_train = cars_shuffled.slice(0, train_count).map(x => x[2]);
var names_test = cars_shuffled.slice(train_count).map(x => x[2]);
return [X_train, X_test, Y_train, Y_test, names_train, names_test];
}
knn = new ML.KNN(data_split[0], data_split[2], {k:5});
knn_results = knn.predict(data_split[1])
knn_accuracy = calculateAccuracy(knn_results, data_split[3])
// A Gaussian Naive Bayes assumes that all variables are Gaussian-distributed for each class.
nb = new ML.NaiveBayes.GaussianNB()
// we have to convert the classes to integers first before we use the Naive Bayes or decision tree classifiers.
classes_to_ids = {
return {'USA': 0, 'Japan': 1, 'Europe': 2};
}
ids_to_classes = {
return {0: 'USA', 1: 'Japan', 2: 'Europe'};
}
nb_results = {
// exercise for the reader?
nb.train(data_split[0], data_split[2].map(x => classes_to_ids[x]));
return nb.predict(data_split[1]);
}
nb_accuracy = calculateAccuracy(data_split[3].map(x => classes_to_ids[x]), nb_results)
// There are a lot of potential options for the decision tree classifier - see https://github.com/mljs/decision-tree-cart
dt = new ML.DecisionTreeClassifier()
dt_results = {
// exercise for reader?
dt.train(data_split[0], data_split[2].map(x => classes_to_ids[x]));
return dt.predict(data_split[1]);
}
dt_accuracy = calculateAccuracy(data_split[3].map(x => classes_to_ids[x]), dt_results)
regression_data_split = {
var cars_data = cars_filtered.map((row) =>
[[row.Cylinders, row.Displacement, row.Horsepower, row.Weight_in_lbs], [row.Miles_per_Gallon], row.Name]);
// note: for MultivariateLinearRegression the output has to be an array
var cars_shuffled = shuffleArray(cars_data);
const train_count = 0.8*cars_shuffled.length; // we will be using 80% of the data for training
var X_train = cars_shuffled.slice(0, train_count).map(x => x[0]);
var X_test = cars_shuffled.slice(train_count).map(x => x[0]);
var Y_train = cars_shuffled.slice(0, train_count).map(x => x[1]);
var Y_test = cars_shuffled.slice(train_count).map(x => x[1]);
var names_train = cars_shuffled.slice(0, train_count).map(x => x[2]);
var names_test = cars_shuffled.slice(train_count).map(x => x[2]);
return [X_train, X_test, Y_train, Y_test, names_train, names_test];
}
lr = new ML.MultivariateLinearRegression(regression_data_split[0], regression_data_split[2])
lr_results = lr.predict(regression_data_split[1])
SS = require('simple-statistics')
// the r^2 function in simple-statistics doesn't actually work for our case so we'll write our own.
r_squared = {
var data_mean = SS.mean(regression_data_split[3].map(x=>x[0]));
var tss = 0;
var res = 0;
for (var i = 0; i < lr_results.length; i++) {
tss += Math.pow(regression_data_split[3][i][0] - data_mean, 2);
res += Math.pow(regression_data_split[3][i][0] - lr_results[i][0], 2);
}
return 1 - res/tss;
}
clustering_data = {
var cars_data = cars_filtered.map((row) =>
[row.Cylinders, row.Displacement, row.Horsepower, row.Weight_in_lbs, row.Acceleration, row.Miles_per_Gallon]);
// note: for MultivariateLinearRegression the output has to be an array
return cars_data;
}
km_results = {
// we will be using 3 clusters for now.
var km = ML.KMeans(clustering_data, 3);
return km;
}
pca = new ML.PCA(clustering_data)
pca_results = pca.predict(clustering_data)
pca_data = {
var d1 = pca_results.getColumn(0);
var d2 = pca_results.getColumn(1);
var temp_data = [];
for (var i = 0; i < cars_filtered.length; i++) {
temp_data.push({'name': cars_filtered[i].Name,
'd1': d1[i],
'd2': d2[i],
'cluster': km_results.clusters[i]});
}
return temp_data;
}
vl.markPoint()
.data(pca_data)
.encode(
vl.x().fieldQ('d1'),
vl.y().fieldQ('d2'),
vl.color().fieldN('cluster'),
vl.tooltip().fieldN('name'),
)
.render()
// There's probably a better way to do this with arquero or something like that
joint_data = {
var temp_data = [];
for (var i = 0; i < cars_filtered.length; i++) {
var c = cars_filtered[i];
c.d1 = pca_data[i].d1;
c.d2 = pca_data[i].d2;
temp_data.push(c);
}
return temp_data;
}
pca.getExplainedVariance()
mobilenet = (await require('@tensorflow-models/mobilenet@0.2')).load()
image_data = Files.url(image_input)
image = {
var image = new Image();
image.src = image_data;
return image
}
mobilenet.classify(image)
toxicity = require('@tensorflow-models/toxicity')
classifier = await toxicity.load(0.9)
// These are sentence examples from a wikipedia comments dataset that was used to train similar models (not this model).
// see: https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/
lines = ["Stop undoing my edits or die!", // a "toxic" comment
"UNBLOCK ME OR I'LL GET MY LAWYERS ON TO YOU FOR BLOCKING MY CONSTITUTIONAL RIGHT TO FREE SPEECH", // another "toxic" comment
"Hi, good day.", // a nontoxic comment
"Thanks for entering this article.", // another nontoxic comment
"idiot did u ever read anything?", // marked as toxic, but try changing the 'o' in 'idiot' to a '0'.
"Considering the vile and perverse anti-Egyptian racism spewed on this talk page on a regular basis, which is all over many Arab countries, especially the Gulf, it is relevant to examine it on an article about Arabs.", // this is marked as a "toxic" comment in the training set.
"Who the hell r u and why do u care neway it was just a bit of fun. Get a life",
"I hate you.", // it's marked as toxic, but are there contexts where this is not toxic?
"I'm working abroad in Australia, and was surprised to find out the green sSkittles over here seem to be Apple flavour.",
"F F F F F F F" // this is a pathological case.
]
results = await classifier.classify(lines)
sentence_results = await classifier.classify([sentence_input])
sentence_toxicity_chance = sentence_results[6].results[0].probabilities[1]
One platform to build and deploy the best data apps
Experiment and prototype by building visualizations in live JavaScript notebooks. Collaborate with your team and decide which concepts to build out.
Use Observable Framework to build data apps locally. Use data loaders to build in any language or library, including Python, SQL, and R.
Seamlessly deploy to Observable. Test before you ship, use automatic deploy-on-commit, and ensure your projects are always up-to-date.