Dynamic optimizer

TensorFlow Special Interest Groups TF.js
, ,
1

Hi! I use tf js, and i have a question
example

        let optimizer = tf.train.adam(0.1, 0.99, 0.999)
        this.model = tf.sequential();
        this.model.add( tf.layers.dense({ inputShape: [units], units: units, activation: 'linear' }));
        this.model.compile({
            optimizer,
            loss: tf.losses.meanSquaredError
        });

How create dynamic optimizer? When i started all work fine, BUT later need more accuracity because I get a lot of losses, not enough accuracy. And if I do high fidelity right away, it takes a VERY long time.
Example

let optimizer = tf.train.adam(0.001, 0.999, 0.9999)

How to make it so that when you get errors - increased accuracy?

3

Hi @gotostereo ,

I apologize for the late response. As per my understanding, TFJS currently does not support creating custom dynamic optimizers. It provides the following built-in optimizers:

tf.train.momentum tf.train.adagrad tf.train.adadelta tf.train.adam tf.train.adamax tf.train.rmsprop

You can select the optimizer that yields the best accuracy and lowest loss for your specific model and dataset.

Let me know if it helps. Thank You!!