NaN Output

[Reichenbachian]

I have a CNN with architecture as shown below.

    model = Sequential()
    model.add(Conv1D(32, 64, strides=2, padding='same', input_shape=(19200, 1))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(MaxPooling1D(pool_size=8))

    model.add(Conv1D(64, 32, strides=2, padding='same'))
    model.add(BatchNormalization())
    model.add(Activation('relu'))
    model.add(MaxPooling1D(pool_size=8))

    model.add(Conv1D(128, 16, strides=2, padding='same'))
    model.add(BatchNormalization())
    model.add(Activation('relu'))

    model.add(Conv1D(256, 8, strides=2, padding='same'))
    model.add(BatchNormalization())
    model.add(Activation('relu'))

    model.add(Flatten())
    model.add(Dense(128))
    model.add(Dropout(p=0.5))

    model.add(Dense(2, name = '...'))
    model.add(Activation('softmax'))
    model.compile(loss='categorical_cross_entropy',
              optimizer='adam',
              metrics='')

The output of the model is [NaN, NaN] regardless of import. The only error output I can find is in the promise stack trace.

Error
    at Promise.longStackTracesCaptureStackTrace [as _captureStackTrace] (http://127.0.0.1:8383/dist/keras.js:22373:19)
    at Promise._then (http://127.0.0.1:8383/dist/keras.js:24303:17)
    at Promise.then (http://127.0.0.1:8383/dist/keras.js:24196:17)
    at <anonymous>:1:105

I have broken it down into the simplest situation possible and am still getting the error. Any suggestions as to what could be causing this?

[Reichenbachian]

The model seems to output correctly, with the output of model.modelDAG being

activation_1: inbound: ["batch_normalization_1"]
layerClass: "Activation"
name: "activation_1"
outbound: ["max_pooling1d_1"]
__proto__: Object
activation_2: {layerClass: "Activation", name: "activation_2", inbound: Array(1), outbound: Array(1)}
activation_3: {layerClass: "Activation", name: "activation_3", inbound: Array(1), outbound: Array(1)}
activation_4: {layerClass: "Activation", name: "activation_4", inbound: Array(1), outbound: Array(1)}
batch_normalization_1: {layerClass: "BatchNormalization", name: "batch_normalization_1", inbound: Array(1), outbound: Array(1)}
batch_normalization_2: {layerClass: "BatchNormalization", name: "batch_normalization_2", inbound: Array(1), outbound: Array(1)}
batch_normalization_3: {layerClass: "BatchNormalization", name: "batch_normalization_3", inbound: Array(1), outbound: Array(1)}
batch_normalization_4: {layerClass: "BatchNormalization", name: "batch_normalization_4", inbound: Array(1), outbound: Array(1)}
conv1d_1: {layerClass: "Conv1D", name: "conv1d_1", inbound: Array(1), outbound: Array(1)}
conv1d_2: {layerClass: "Conv1D", name: "conv1d_2", inbound: Array(1), outbound: Array(1)}
conv1d_3: {layerClass: "Conv1D", name: "conv1d_3", inbound: Array(1), outbound: Array(1)}
conv1d_4: {layerClass: "Conv1D", name: "conv1d_4", inbound: Array(1), outbound: Array(1)}
dense_1: {layerClass: "Dense", name: "dense_1", inbound: Array(1), outbound: Array(1)}
dropout_1: {layerClass: "Dropout", name: "dropout_1", inbound: Array(1), outbound: Array(1)}
flatten_1: {layerClass: "Flatten", name: "flatten_1", inbound: Array(1), outbound: Array(1)}
gender_cnn_activation: {layerClass: "Activation", name: "gender_cnn_activation", inbound: Array(1), outbound: Array(0)}
gender_cnn_output: {layerClass: "Dense", name: "gender_cnn_output", inbound: Array(1), outbound: Array(1)}
input: {layerClass: "InputLayer", name: "input", inbound: Array(0), outbound: Array(1)}
max_pooling1d_1: {layerClass: "MaxPooling1D", name: "max_pooling1d_1", inbound: Array(1), outbound: Array(1)}
max_pooling1d_2: {layerClass: "MaxPooling1D", name: "max_pooling1d_2", inbound: Array(1), outbound: Array(1)}

[Reichenbachian]

Isolated the problem layer using brute force. It's the MaxPooling1D layer. Probably #79, repeated.

[Reichenbachian]

Found the bug! It was same as #79. I rewrote the _Pooling1D.js It no longer outputs NaN. Here are the changes...

/**
 * _Pooling1D layer class
 */
export default class _Pooling1D extends Layer {
  /**
   * Creates a _Pooling1D layer
   */
  constructor(attrs = {}) {
    super(attrs)
    this.layerClass = '_Pooling1D'

    const { pool_size = 2, strides = null, padding = 'valid' } = attrs
    if (Array.isArray(pool_size)) {
      this.poolSize = pool_size[0]
    } else {
      this.poolSize = pool_size
    }
    if (Array.isArray(strides)) {
      this.strides = strides[0]
    } else if(strides === null) {
      this.strides = this.poolSize
    } else {
      this.strides = strides
    }
    // this.strides = strides === null ? this.poolSize : strides
    this.padding = padding

    // default pooling function
    // can be `max` or `average`
    this.poolingFunc = 'max'
  }