Learning aggregators using neural networks in TensorFlow 2.0

TensorFlow General Discussion
1

I am trying to design the following neural network model using Tensorflow:

One of the model’s inputs is X, a list of n vectors of dimension 3. The second input to the model is Y, a list of n natural numbers in ascending order starting with 0. The model’s output is Z, a list of m vectors of dimension 3.

There are m unique numbers in Y representing the class of input vectors of dimension 3. The number of input vectors of different classes may differ.

The first layer of the model transforms each vector in X to a vector of dimension 2 and applies the ‘gelu’ activation function. The second layer performs ‘segment_sum’ to reduce n vectors of dimension 2 to m vectors of dimension 2 using Y. The third layer transforms m vectors of dimension 3, which is the model’s output.

I use cosine dissimilarity loss and Adam optimizer to train the model.

Here is the code that I wrote for this purpose:

import numpy as np
import tensorflow as tf
from tensorflow import keras

# Prepare the input and output data (example)
n = 10
m = 4
X = np.random.random((n, 3)).astype('float32')
Y = np.array([0, 0, 1, 1, 2, 2, 2, 3, 3, 3]).astype('int32')
Z = np.random.random((m, 3)).astype('float32')

class CustomModel(tf.keras.Model):
    def __init__(self):
        super(CustomModel, self).__init__()
        self.dense1 = keras.layers.Dense(2, activation='gelu')
        self.dense2 = keras.layers.Dense(3)

    def call(self, inputs):
        X, Y = inputs
        X = self.dense1(X)
        X = tf.math.segment_sum(X, Y)
        Z = self.dense2(X)
        return Z


model = CustomModel()

model.compile(loss=tf.keras.losses.CosineSimilarity(axis=1), optimizer=tf.keras.optimizers.Adam())

model.fit([X, Y], Z, epochs=10)

The model is essentially designed to learn an aggregation function. However, I get the following error:

Traceback (most recent call last):
  File "/home/nitesh/PycharmProjects1/pythonProject/research/reasoning_with_vectors/custom_model.py", line 31, in <module>
    model.fit([X, Y], Z, epochs=10)
  File "/home/nitesh/miniconda3/envs/relbert/lib/python3.10/site-packages/keras/utils/traceback_utils.py", line 70, in error_handler
    raise e.with_traceback(filtered_tb) from None
  File "/home/nitesh/miniconda3/envs/relbert/lib/python3.10/site-packages/keras/engine/data_adapter.py", line 1852, in _check_data_cardinality
    raise ValueError(msg)
ValueError: Data cardinality is ambiguous:
  x sizes: 10, 10
  y sizes: 4
Make sure all arrays contain the same number of samples.

I tried a lot, but I didn’t find any way to encode the model using TensorFlow 2.0. Could anyone help? Thanks in advance.

3

Hi @Nitesh_Kumar,

Sorry for the delayed response. Data Cardinality error is due to mismatch with the number of samples in X, Y and target Z. X is having 10 samples, Y seems to have 10 samples but when Y is passed to `tf.math.segment_sum(X, Y), Y has segment_Ids length 4. The target Z should have maintain the same number samples as X. So Please modify the input according to your Custom layer. Here is the sample code to maintain the data cardinality among X, Y and Z.

n = 10  
m = 10  
X = np.random.random((n, 3)).astype('float32')  
Y = np.array([0, 0, 1, 1, 2, 2, 2, 3, 3, 3]).astype('int32')  
Z = np.random.random((m, 3)).astype('float32')  and Y)

If still facing issue with cardinality, please let us know your response for further assistance.

Thank You