Applied Machine Learning - Week 4

Monday the 17th - Friday the 21rd of May 2021

Monday 17th of May (afternoon):
Lectures: Recurrent Neural Networks (RNN), Long Short Term Memory (LSTM), and Backpropagation (TP).
     The ImageNet Competition - a revolution in ML (TP).

Zoom:Link to lecture.
     Recording in Lecture video I (222 MB), Lecture video II (76 MB), and Lecture chat (2 kB).

Exercise: Exercise: Using an RNN/LSTM/GRU, predict the next entries in a sinus (periodic) and Mackay (non-periodic) sequence: SequenceTraining.ipynb.
     Once you become familiar with LSTMs, try to make FlightPassengerPredictions.ipynb on 1949-1961 airline data.
     Also, the exercise will be used for coordination/group discussion of final project.


Wednesday 19th of May (morning):
Lectures: An introduction to Convolutional Neural Networks (Aleksandar Topic).
     Very cool visualisation of how a CNN works.

Zoom:Link to lecture (Recorded!).
     Recording in Lecture video (112 MB).

Exercise: See if you can recognise handwritten numbers with a Convolutional Neural Network: CNN_MNISTdata.ipynb.
     The data can (also) be obtained from (famous) Yann LeCun's webpage: Link to MNIST dataset.


Wednesday 19th of May (afternoon):
Lectures: Graph Neural Networks - analysing geometric data (Rasmus Oersoe).

Zoom:Link to lecture (Recorded!).
     Recording in Lecture video (78 MB).

Exercise: We will use the afternoon to work on the Small project, which is due by Monday the 24th of May.



Housing cleaning, clustering, and estimating exercise:
The housing data consists of about 50000 housing sales, where 90+ features are provided along with the actual sales price. The original data HousingPrices_Org.csv (21 MB) first needs cleaning. Following cleaning, one can add features (here GPS coordinates and distance to sea) using the additional data files:
  • GPS_data.csv (1.3 MB)
  • SEA_DIST.csv (680 kB)
    Example code for doing this can be found here:
  • RegressionOnHousing_Clean_Data.py (3.7 kB)
  • RegressionOnHousing_Feature_Adding.py (11 kB)
    An example of the resulting data file can be found here: HousingPrices_Cleaned.csv (8.8 MB), and example code for actual sales price estimates here: RegressionOnHousing.py (34 kB).
    Last updated: 10th of May 2021 by Troels Petersen.