Rostlab / DM_CS_WS_2016-17

Repo for general info of the course and communication
21 stars 5 forks source link

Feedback due for Jan 25th #23

Closed lrrichter closed 7 years ago

lrrichter commented 7 years ago

Terrorism: You could try to convert/normalize the victim count to a relative accuracy. +/- 1 definitely matters much more with 0 or 1 victim than with 100 cases. That corresponds to relative fluctuations between 100% and 1%.

Food Facts: 1.1 In the figures what is the y-axis for? 1.2 it would be good to know the base distribution of the (test) data, i.e total # of instances, how many positives etc... 1.3 also give base values

Google N-Grams Bad luck, that the specific play prediction with reduced sample do not work. Is there a threshold were prediction become really precise, or a plateau? Are there already results for the 2-step approach? BTW good work.

UK-Crime For performance assessment: sometime relative / normalized values make it easier to compare the overall performance Also you could make more out of your figures if you would add some explanations and interpretations

Global climate Good work, even if some approaches did not work so far. Main observation: Temperatures are hardly measured to 1/100 degrees and you values with an accuracy of up to 1/1000000 degree. This may be interpreted as completely clueless about the meaning (which - I think - is not the case).

London Real Estate Market Not so much to review yet. Can you estimate if there is an intrinsic difference in apartments for sale in comparison to apartment for rent? May be size, location, price, etc...?