Question

We are interested in making a neural network that can predict the next element in a sequence. For each of the sequences below, indicate all of the following models that could make correct predictions if its weights (and in case of the recurrent model, the dimension d ) were set correctly don't worry about learning. The possible models are:

A) A bigram model, which is a feedforward network that takes sequence element yₜ−1​ as input and generates element yₜ​ as output.
B) A trigram model, which is a feedforward network that takes sequence elements yₜ−2​ and yₜ −1​ as input and generates element yₜ as output.
C) A simple recurrent model, with state s t​ of dimension d×1, governed by the equations: sₜ​ = tanh(W ss s t−1 +W sx yt−1​ ) and y t​ =softmax(W o s t​ ) The sequence is 1,1,−1,1,1,−1,1,1,−1,…, which model(s) work?

Answer 1

Using the regression model, the predicted sales for day 60 would be 250.12 thousand dollars and for day 90 would be 324.52 thousand dollars.

Step-by-step explanation:

To predict the sales growth for an electronics retailer, we can use the provided linear regression model ŷ = 101.32 + 2.48x, where ŷ represents the predicted sales in thousands of dollars and x is the day within the first quarter. For day 60, we substitute x with 60 to get ŷ = 101.32 + 2.48(60) which equals 101.32 + 148.8, resulting in a prediction of 250.12 thousand dollars. Similarly, for day 90, the prediction would be ŷ = 101.32 + 2.48(90), which equals 101.32 + 223.2, resulting in a prediction of 324.52 thousand dollars.