Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Document Layer.apply() kwargs options

See original GitHub issue

TensorFlow.js version

tfjs-core-0.15.2

Browser version

N/A

Describe the problem or feature request

When building a Layers model with a preLU, reLU, or leakyReLU layer, optimizing using the core API throws an error.

Code to reproduce the bug / link to feature request

Here’s the code that builds a simple Layers model with an activation function. Using ELU as activation function throws no error (having no activation function works fine, as well). On the contrary, PreLU, reLU, and leakyReLU fail.

const tf =  require('@tensorflow/tfjs-node');
const model = tf.sequential();
model.add(tf.layers.dense({ inputDim: 1, units: 1 }));
model.add(tf.layers.leakyReLU()); // Fails
// model.add(tf.layers.prelu()); // Fails
// model.add(tf.layers.reLU()); // Fails
// model.add(tf.layers.elu()); // Works

const xs = tf.tensor2d([[Math.random()], [Math.random()]]);
const ys = tf.tensor2d([[Math.random() + 10], [Math.random() + 10]]);
tf.train.sgd(0.1).minimize(() => model.apply(xs).sub(ys).square().mean());

Output when using prelu:

2019-03-18 15:00:36.251109: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX
/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:107
            throw ex;
            ^

Error: Tensor is disposed.
    at Tensor.throwIfDisposed (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tensor.js:284:19)
    at Tensor.greater (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tensor.js:585:14)
    at grad (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/ops/relu_ops.js:59:23)
    at Object.tapeNode.gradient (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:144:45)
    at _loop_1 (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tape.js:84:35)
    at Object.backpropagateGradients (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tape.js:112:9)
    at /home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:363:20
    at /home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:91:22
    at Engine.scopedRun (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:101:23)
    at Engine.tidy (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:90:21)

Output when using reLU:

2019-03-18 15:02:00.429530: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX
/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:107
            throw ex;
            ^

Error: Tensor is disposed.
    at Tensor.throwIfDisposed (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tensor.js:284:19)
    at Tensor.step (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tensor.js:775:14)
    at grad (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/ops/relu_ops.js:17:26)
    at Object.tapeNode.gradient (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:144:45)
    at _loop_1 (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tape.js:84:35)
    at Object.backpropagateGradients (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tape.js:112:9)
    at /home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:363:20
    at /home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:91:22
    at Engine.scopedRun (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:101:23)
    at Engine.tidy (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:90:21)

Output when using leakyReLU:

/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:107
            throw ex;
            ^

TypeError: Cannot read property 'values' of undefined
    at NodeJSKernelBackend.getInputTensorIds (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-node/dist/nodejs_kernel_backend.js:142:26)
    at NodeJSKernelBackend.executeSingleOutput (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-node/dist/nodejs_kernel_backend.js:186:73)
    at NodeJSKernelBackend.greaterEqual (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-node/dist/nodejs_kernel_backend.js:429:21)
    at environment_1.ENV.engine.runKernel.$a (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/ops/compare.js:89:83)
    at /home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:129:26
    at Engine.scopedRun (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:101:23)
    at Engine.runKernel (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/engine.js:127:14)
    at greaterEqual_ (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/ops/compare.js:89:37)
    at Object.greaterEqual (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/ops/operation.js:23:29)
    at Tensor.greaterEqual (/home/ubuntu/issue/node_modules/@tensorflow/tfjs-core/dist/tensor.js:594:26)

Issue Analytics

State:
Created 5 years ago
Comments:5 (4 by maintainers)

Top GitHub Comments

1reaction

dsmilkovcommented, Mar 19, 2019

I just manually verified that this is fixed in https://github.com/tensorflow/tfjs-core/pull/1604 which will come in the next release 1.0.2 by end of week.

This means you don’t need to pass {training: true} in the next release.

Thanks for the feedback and the great bug reproduce @josei !

0reactions

dsmilkovcommented, Mar 19, 2019

@caisq Does Python Keras have training=true as an argument when calling a layer functionally, or is this specific to TF.js? If specific to TF.js, curious when did this option to apply() got added? Thanks!

Top Results From Across the Web

Proper way to use **kwargs in Python - Stack Overflow

I like to use positional arguments only for required arguments, and kwargs for arguments that may or may not be specified, but it...

Layer weight regularizers - Keras

Layer weight regularizers. Regularizers allow you to apply penalties on layer parameters or layer activity during optimization. These penalties are summed ...

tf.keras.layers.Layer | TensorFlow v2.11.0

call(self, inputs, *args, **kwargs) : Called in __call__ after making sure build() has been called. call() performs the logic of applying the layer...

Folium 0.12.1 documentation

For more advanced tile layer options, use the TileLayer class. ... **kwargs – Additional keyword arguments are passed to Leaflets Map class: ...

API documentation - gmaps figure

You can use the drawing layer to add lines, markers and polygons to a map: >>> fig = gmaps.figure() >>> drawing = gmaps.drawing_layer(features=[...