neural network - Tensorflow autoencoders with one vector per input neuron -
i'm new tensorflow , deep neural networks. i'm trying anomaly detection on trajectories using autoencoders , i'm having issue model.
i'm not able right weight matrix / not sure how it.
here model:
- each input neuron of encoder receive vector 4 features (this vector correspond observation part of trajectory).
- the number of input neurons correspond number of observation (which 289).
- i have total of 336 trajectories correspond batch
therefore input data shape (336,289,4)
- i have 2 hidden layers; on each 1 divide number of previous neurons 2 h1 have 144 neurons , h2 72 neurons
for weights, have:
weights = { 'encoder_h1': tf.variable(tf.random_normal([336, n_hidden_1, 289])), 'encoder_h2': tf.variable(tf.random_normal([336, n_hidden_2, n_hidden_1])), 'decoder_h1': tf.variable(tf.random_normal([336, n_hidden_1, n_hidden_2])), 'decoder_h2': tf.variable(tf.random_normal([336, n_input, n_hidden_1 ])), }
and activation function sigmoid
tf.nn.sigmoid(tf.add(tf.matmul(weights['encoder_h1'],x), biases['encoder_b1'])
but i'm afraid gives wheight matrix trajectory or want weight matrix trajectories, should 2d tensor don't know how proceed.
i tried many thing such removing 336 part weight shape tensorflow says tha not possible matmul on 3d , 2d tensor.
do have idea on how do?
thanks in advance help
Comments
Post a Comment