Computer science major master student interested in computer vision and deep learning
Published
Edited
May 7, 2018
3 forks
23 stars
tf = require('@tensorflow/tfjs@0.8.0')
show(tf.scalar(3))
show(tf.tensor([2,2]))
show(tf.zeros([2,2]))
show(tf.tensor([2,3,4]).square())
show(tf.tensor([1,2,3]).square().square())
{
let x = tf.tensor([1,2,3]);
x.dispose()
return x
}
function f(x)
{
return tf.tidy(()=>{
const y = x.square();
const z = x.mul(y);
return z
});
}
model = {
//create a sequential method
const model = tf.sequential();
//add frist conv layer
model.add(tf.layers.conv2d({
inputShape: [28, 28, 1],
kernelSize: 5,
filters: 8,
strides: 1,
activation: "relu",
kernelInitializer: "VarianceScaling"
}));
//add first pooling layer
model.add(tf.layers.maxPooling2d({
poolSize: [2, 2],
strides: [2, 2]
}));
//add second conv layer
model.add(tf.layers.conv2d({
kernelSize: 5,
filters: 16,
strides: 1,
activation: "relu",
kernelInitializer: "VarianceScaling"
}));
//add the second pooloing layer
model.add(tf.layers.maxPooling2d({
poolSize: [2, 2],
strides: [2, 2]
}));
//flatten the model to prepare the input for the dense layer
model.add(tf.layers.flatten());
//a dense layer with output equals the number of classes
model.add(tf.layers.dense({
units: 10,
kernelInitializer: "VarianceScaling",
activation: "softmax"
}));
//compile the model using and adam optimizer
await model.compile({
optimizer: tf.train.adam(0.0001),
loss: "categoricalCrossentropy",
metrics: ["accuracy"]
});
//return the model
return model;
}
function createBatch(BATCH_SIZE)
{
let input = tf.zeros([BATCH_SIZE,28,28,1]);
let labels = tf.oneHot(tf.zeros([BATCH_SIZE]),10)
return [input , labels]
}
batch = createBatch(5)
batch[0].shape
show(batch[1])
h = await model.fit(batch[0], batch[1],
{
batchSize: 10,
epochs: 1
})
show(model.predict(tf.zeros([1, 28, 28, 1])))
loss = h.history.loss[0]
accuracy = h.history.acc[0]
mobilenet.inputs[0].shape
function preprocess(img)
{
let tensor = tf.fromPixels(img).toFloat();
const offset = tf.scalar(127.5);
// Normalize the image
const normalized = tensor.sub(offset).div(offset);
//We add a dimension to get a batch shape [1,224,224,3]
const batched = normalized.expandDims(0)
return batched
}
preImg = preprocess(img)
function predict(input)
{
//get predictions
let pred = mobilenet.predict(input);
//retreive the highest probability class label
let idx = pred.argMax().buffer().values[0];
return idx
}
cls = predict(preImg)
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.
