Hierarchical classification using LCPN (Local Classifier per Parent Node) approach - tensorflow

Question

Objective:
I am working on a Hierarchical Classification problem and want to solve it using the Local Classifier per Parent Node (LCPN) approach using Tensorflow. In order to do so, I have to create local classifiers based on the hierarchical dataset.
For example:
I have manually created a hierarchical tree structure for the CIFAR-10 Dataset following this paper. The hierarchical structure is as follows:
Based on this structure, it requires a total of 6 local classifiers.
1 Classifier in level 1:
for classifying class transport and animal.
2 Classifier in level 2:
1 for classifying classes sky, water, road (Subclass of class transport )
1 for classifying classes bird, reptile, per, medium (Subclass of class animal)
3 Classifier in level 3:
1 for classifying classes automobile and truck (Subclass of the class road)
1 for classifying classes cat and dog (Subclass of class pet)
1 for classifying classes deer and horse (Subclass of class medium)
NOTE:
I want to get all the predictions on level 3 (10 classes). If the classifier for level 1 outputs a class of level 2 that does not have more than one subclass in level 3, it should automatically assign the corresponding class in level 3 for that sample.
For example: If the 1st classifier identifies a sample as transport, then it will select the classifier to classify the subclass of transport (sky, water, road). If the classifier in level 2 classifies that sample as sky then will no longer need another classifier to classify the subclass as it has only one subclass which is the class airplane. But for my implementation, I want the final prediction as level 3 predictions and output as airplane.
Implementation:
To implement this, so far I have done the following:
I have determined the number of local classifiers from the dataset and the number of classes by using treelib. It determines the number of outputs requires for the local classifiers.
I am working on generating a dataset pipeline using tf.data.Dataset.filter which will provide a filtered dataset for training the models. As I am going to train the local classifier with relevant samples. For example, the classifier for determining the subclass of level 1 class transport will be trained with samples of all the classes that are under the level 1 class transport. So, I want to filter out the samples that belong to the class animal or any subclass of animal.
After that, I have to implement a decision tree for predicting from the models.
Now, I am struggling with the implementation using this approach. Is there any better solution for this kind of problem? Or any alternative approach?

Answer 1

Look at it this way when you create an initial of information with 10 samples in one input, you do it the same way and capture the results in each layer.
You know about the classification problems you do it by finding maximum or minimum, ranges or critical.
Why do the notebooks select LAN interfaces over WiFi? Speed and Priorities even some of them are faster than LAN.
Sample: Simply print out each time you had herachy.
import tensorflow as tf
class MyDenseLayer(tf.keras.layers.Layer):
def __init__(self, num_outputs):
super(MyDenseLayer, self).__init__()
self.num_outputs = num_outputs
def build(self, input_shape):
self.kernel = self.add_weight("kernel",
shape=[int(input_shape[-1]),
self.num_outputs])
def call(self, inputs):
return tf.matmul(inputs, self.kernel)
start = 3
limit = 33
delta = 3
# Create DATA
sample = tf.range(start, limit, delta)
sample = tf.cast( sample, dtype=tf.float32 )
# Initail, ( 10, 1 )
sample = tf.constant( sample, shape=( 10, 1 ) )
layer = MyDenseLayer(10)
data = layer(sample)
# Layer 1, ( 10, 2 )
layer = MyDenseLayer(2)
data = layer(data)
# Layer 2, ( 10, 7 )
layer = MyDenseLayer(7)
data = layer(data)
# Layer 3, ( 10, 10 )
layer = MyDenseLayer(10)
data = layer(data)
print( data )