Evaluation of User defined cost function

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

Hi Everyone!
Q.1. In ANN, we train our neural network with input data and targeted data… like house size and price, where house size is input data and house price is targeted data.
Q.2. how we can achieve optimized output from ANN with respect to specific feature.? like in house price and size example we train out model with price and size. Is it possible that I input price and ANN evaluate it with respect to “lowest age and number of bedrooms” and return me the best possible available price with respect to above two feature? How I can do this?

2

Hi, @Muhammad_Azhar_Ghaur !

Sure! You can input as many features as you want. You just have to concatenate the input features along a vector and make sure the network has the same input dimension.

3

Hello @Muhammad_Azhar_Ghaur

We train a model with n features to predict the target y. Once the training is over, we can use the trained model to make predictions.

To make predictions with the trained model, we will need to provide the value of the n input features (as many input features as was used during training) and the model will then return a predicted value \hat{y}.

So, if your question is:

  1. whether we can provide a few of the input features to the model and the model will then fill in the values of the remaining missing features with optimal values, the answer is NO.
  2. if we can provide the value of the target variable, price, in this case and the model will return the values of the input features age and number of bedrooms - The answer is again NO.
1 Like
4

Now its very clear to me…So if we need to predict something in accordance to some specific feature we need to train our model with those feature. In this way only we can get optimized output from ANN…

5

That is correct @Muhammad_Azhar_Ghaur

2 Likes
6

I have one more question. I f we are normalizing a data like by using Z-score normalization than how can we perform de-normalization to get the desired value? like suppose we normalized to house price (between o.1 million to 1 million) between 0.1 to 1 and at the output wee need price in real number like 500,000. than how can we achieve it ?