Project

Part 1:

What type of preprocessing was applied to the raw input data? Why was it necessary to do so?
Does your model suffer from overfitting? If so, what would you suggest doing about it?
Is accuracy a good metric of success in this case? Why (not)? Ams:

Part 2:
What type of preprocessing was applied to the raw input data? Why was it necessary to do so?
How many categories are there in this case?
Does your model suffer from overfitting? If so, what would you suggest doing about it?
Is accuracy a good metric of success in this case? Why (not)?

Part 3:
What type of preprocessing was applied to the raw input data? Why was it necessary to do so?
Why is this problem a case of regression(rather than classification)?
Does your model suffer from overfitting?If so, what would you suggest doing about it?
Is mean absolute error(MAE)a good metric of success in this case? Why (not)?

TODO 1
"Deep Learning with Python" by François Chollet (for a practical introduction to deep learning and its applications)
"Sentiment Analysis with Deep Learning using the IMDB dataset" (for binary classification of movie reviews)
"Text Classification with TensorFlow and Keras using the Reuters dataset" (for multiclass classification of newswires)
"Predicting House Prices with Regression using the Boston Housing Price dataset" (for regression for house price estimation)

TODO 3
Write code to display 2 cases (one false positive, one false negative) where the classifier makes mistakes. Make sure to display both the true value as well as the predicted value.

Comment on the mistakes and what you believe might have caused each of them.

TODO 4
Write code to try different combinations of:

numbers of hidden layers and units per layer
loss functions
activation functions

Please refrain from using better architectures (RNN, LSTM, etc.) and "advanced" techniques to curb overfitting (e.g., dropout, batch normalization, weight regularization, etc.)

TODO 5
Write code to try different combinations of:

numbers of hidden layers and units per layer

Please refrain from using better architectures and "advanced" techniques to curb overfitting (e.g., dropout, batch normalization, weight regularization, etc.)

mport matplotl b.pyplot as plt mport keras
from keras mport layers

(OPTIONAL) TODO 6
Write code to try different combinations of:

K for K-fold validation
number of epochs
number of units per hidden layer
number of hidden layers

Please refrain from using better architectures and "advanced" techniques to curb overfitting (e.g., dropout, batch normalization, weight regularization, etc.)