Public
Edited
May 11, 2023
2 stars
model_dict = {
reset_parameters;
// Feature weights and bias - the parameters we want to learn
const w = tf.tidy(
() => tf.variable(tf.randomNormal([D,1], 0, 1.0), true)
);
const b = tf.tidy(
() => tf.variable(tf.randomNormal([1], 0, 1.0), true)
);
// Logistic regression model: P(y=1 | x) = Bernoulli(y=1 | sigm(w.T * x + b))
// where * is matrix multiplication,
// w, x ∈ ℝ^D, b ∈ ℝ, and y ∈ {0,1}
const f = x =>
x.matMul(w)
.add(b)
.sigmoid()
.as1D();
yield {'model': f, 'w': w, 'b':b};
try {
yield invalidation;
} finally {
w.dispose();
b.dispose();
}
}
loss = (pred, truth) => tf.metrics.binaryCrossentropy(truth, pred)
current_loss = {
step; // Take an optimization step when the button is clicked
const optimizer = tf.train.sgd(learningRate);
// What parameters are we optimizing?
const vars = [];
if (params_to_optimize.includes('w'))
vars.push(model_dict.w);
if (params_to_optimize.includes('b'))
vars.push(model_dict.b);
// Optimize the loss or evaluate it if no parameters selected
let current_loss = 0;
if (vars.length > 0)
current_loss = tf.tidy(
() => optimizer.minimize(
() => loss(model(X), y), true, vars
)
);
else
current_loss = tf.tidy(
() => loss(model(X), y)
);
// Yield result and clean up
optimizer.dispose();
yield current_loss;
try {
yield invalidation;
} finally {
current_loss.dispose();
}
}
import {latex_matrix} from "@sorig/latex_matrix"
import {button, slider, checkbox} from "@jashkenas/inputs"
tick = {
while (true) {
yield Promises.tick(1000);
}
}
Purpose-built for displays of data
Observable is your go-to platform for exploring data and creating expressive data visualizations. Use reactive JavaScript notebooks for prototyping and a collaborative canvas for visual data exploration and dashboard creation.
