tarunlnmiit / data-mining

It's been fun guys :) Take care and good luck for the exam!

One more thing: Learn how to read set notation. Some questions in the exam might be formed using set notation (like in the slides) instead of being given in English, so watch a youtube video on how to read set notation or something. Also, feel free to email me with questions. I know I skipped describing the itemset lattice thing, so if you don't understand it, just email me and I'll tell you how to construct it.

There was a mistake in the lattice in this notebook https://github.com/tchanda90/data-mining/blob/master/notebooks/frequent_itemsets.ipynb The itemset abde was marked as I (pruned because it doesn't meet minimum support count), but it should have been marked P (pruned because a subset is infrequent). Fixed now.

Download this notebook to generate cluster heatmaps https://github.com/tchanda90/data-mining/blob/master/notebooks/cluster_heatmaps_and_validation.ipynb

Uploaded iso accuracy lines https://github.com/tchanda90/data-mining/blob/master/notebooks/iso_accuracy_lines.ipynb

Vote for questions/concepts to be discussed on the last day. You may include questions from the lecture, exercises, or the assessment tests. https://forms.gle/s8AEVEcpkPDpW9iM9

Uploaded exercise 9 https://github.com/tchanda90/data-mining/blob/master/notebooks/finding_optimal_k.ipynb

Uploaded exercise 8. For the calculation of cohesion and separation, follow the steps shown in the notebook. Please do not normalize the distances. Follow the exact formulae and steps shown in the notebook https://github.com/tchanda90/data-mining/blob/master/notebooks/cluster_validation.ipynb The values are confirmed correct.

IMPORTANT: For the computation of the Silhouette coefficient in task 9.1, please calculate b as average distance to the points of the nearest cluster (just like in the book) and not as average distance to the points of all other clusters as shown in the slides. This is confirmed to be correct. https://github.com/tchanda90/data-mining/blob/master/notebooks/silhouette_score.ipynb ALSO, please calculate the overall clustering silhouette coefficient as average over all point silhouette values, as shown here in the notebook. In the slides, the overall clustering silhouette coefficient is calculated as average cluster silhouette coefficient which doesn't take different cluster sizes into account.

If you have any doubts regarding the above, feel free to email me [email protected]