How do I use a masking layer for TCN? I want to mask certain time steps which are missing. about keras-tcn HOT 4 CLOSED

arvind267 commented on May 25, 2024

How do I use a masking layer for TCN? I want to mask certain time steps which are missing.

from keras-tcn.

Comments (4)

philipperemy commented on May 25, 2024

Like this: https://keras.io/api/layers/core_layers/masking

import numpy as np
import tensorflow as tf

from tcn import TCN

samples, timesteps, features = 32, 10, 8
inputs = np.random.random([samples, timesteps, features]).astype(np.float32)

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Masking(mask_value=0., input_shape=(timesteps, features)))
model.add(TCN(32))

output = model(inputs)
print(output)

from keras-tcn.

philipperemy commented on May 25, 2024

I'm adding more comments here. Might be useful for later. It seems to work correctly because:

inputs[:, :, :] = 1.
model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Masking(mask_value=1., input_shape=(timesteps, features)))

The model will output only zeros. If we change it to inputs[:, :, :] = 2., the model will output different values.

Another example:

import numpy as np
import tensorflow as tf

from tcn import TCN

samples, timesteps, features = 32, 10, 8
inputs = np.random.random([samples, timesteps, features]).astype(np.float32)
inputs[:, :-1, :] = 1.

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Masking(mask_value=1., input_shape=(timesteps, features)))
model.add(TCN(16, return_sequences=True))

output = model(inputs)
print(output.shape)
print(np.mean(np.abs(output[:, :-1, :]))) # 0.0 EXPECTED (EQUAL TO 0, due to masking).
print(np.mean(np.abs(output[:, -1:, :]))) # 0.8137888 EXPECTED (DIFFERENT THAN 0).