Categorical Variable Reduction using NN

Question

I was trying to categorical variable engineering following this paper. The code is the following:

import random
import pandas
import numpy as np
import tensorflow as tf

from tensorflow.contrib import layers
from tensorflow.contrib import learn
from __future__ import print_function

from sklearn.preprocessing import LabelEncoder

My dataset looks like the following. It's has 2 independent variable ('X1' & 'X2')and 1 dependent variable ('lable'). 'X2' is the categorical variable. I want to create an embedding vector for this variable and run the simple linear regression to predict 'label'using Tensorflow. I could use any other method. But since linear regression is easiest to understand, I'm trying that.

df = pd.DataFrame({'X1': np.array(["A","A","B","C","B","C","B","C","C","B",
                         "A","B","A","C","A","A","C"]),'X2': np.array([3.3,4.4,5.5,6.71,6.93,4.168,9.779,6.182,7.59,2.167,
                         7.042,10.791,5.313,7.997,5.654,9.27,3.1]),
                       'label': np.array([1.7,2.76,2.09,3.19,1.694,1.573,3.366,2.596,2.53,1.221,
                         2.827,3.465,1.65,2.904,2.42,2.94,1.3])})

For variable 'X1', I'm creating levels.

encoder = LabelEncoder()
encoder.fit(df.X1.values)
X = encoder.transform(df.X1.values)

Recreating dependent variable list.

y = np.asarray([1.7,2.76,2.09,3.19,1.694,1.573,3.366,2.596,2.53,1.221,
                         2.827,3.465,1.65,2.904,2.42,2.94,1.3])

Setting Hyper-parameters

training_epochs = 5
learning_rate = 1e-3
cardinality = len(np.unique(X))
embedding_size = 2
input_X_size = 1
n_hidden = 10

Setting up variables:

embeddings = tf.Variable(tf.random_uniform([cardinality, embedding_size], -1.0, 1.0))


h = tf.Variable(tf.truncated_normal((embedding_size + len(df.X1), n_hidden), stddev=0.1))


W_out = tf.get_variable(name='out_w', shape=[n_hidden],
                            initializer=tf.contrib.layers.xavier_initializer())

Embedding:

embedded_chars = tf.nn.embedding_lookup(embeddings, x)
embedded_chars = tf.reshape(embedded_chars, [-1])
embedded_chars= embedded_chars + np.array([3.3,4.4,5.5,6.71,6.93,4.168,9.779,6.182,7.59,2.167,
                         7.042,10.791,5.313,7.997,5.654,9.27,3.1])

Multiplying with Hidden Layers:

layer_1 = tf.matmul(embedded_chars,h)
layer_1 = tf.nn.relu(layer_1)
out_layer = tf.matmul(layer_1, W_out)

# Define loss and optimizer

cost = tf.reduce_sum(tf.pow(out_layer-y, 2))/(2*n_samples)
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

Run the graph

init = tf.global_variables_initializer()

# Launch the graph
with tf.Session() as sess:
    sess.run(init)

    for epoch in range(training_epochs):
        avg_cost = 0.

        _, c = sess.run([optimizer, cost],
                        feed_dict={x: X, y: Y})
print("Ran without Error")

While running the code, I'm getting the following error.

ValueError: Shape must be rank 2 but is rank 1 for 'MatMul_1' (op: 'MatMul') with input shapes: [17], [19,10].

I'm not able to add the continuous variable with embedding variable.

Can anyone please guide me how to do it?

Thank you!

Answer 1

You are multiplying matrices, so the dimensions have to match up.

The following line reshapes the embedding layer to a rank 1 matrix, hence causing the error.

embedded_chars = tf.reshape(embedded_chars, [-1])

You may want to rethink the embedding portion of the code (the function's docstring), as that is what needs to be understood perfectly in order to define the computational graph.