Luise - DREU Research Intern
As I mentioned last week, we decided to increase the size of our dataset hoping that it would improve our model. After finishing the training, we did see small improvements, but there is still room for more! While I have been blaming the data, I found a pretty bad bug earlier this week that might be part of the reason behind the not so convincing results. Basically, an important part of the training, is to check if each word in the dataset’s transcripts is in the lexicon we train with. In general too many of the so called “out of vocabulary” words (OOVs) are not great in the training data. What I realized was that the lookup didn’t ignore case and punctuation - so neither “It” nor “it.” would be recognized even if the word “it” was in the lexicon. I won’t go more into detail, but basically I hope that fixing this bug will lead to much better results when I finish training next week. While I last week increased the amount of data, I this week worked on decreasing datasets and training on them. It will be interesting to see how the changing amount of data affects the results.
This week we also started “Lab Talks.” Every Wednesday for the rest of the summer we (all of the people in my lab) will meet over cookies and coffee and one of us will give an informal presentation on what we have been up to, our frustrations, solutions and so on.
This Saturday I walked all the way down to SoHo - which is around 6-7 miles away, I think - and spent the afternoon there. I love walking in the city because I really get to see its different parts, small shops and nice cafes. Another nice place I have discovered is the Hudson River Greenway. On one of my runs in Riverside Park, I accidentally ended on a path that lead me right to the Greenway, and ever since I have intentionally aimed towards it. The Greenway goes along the Hudson River on the West side and then all around Manhattan. The parts close to campus are literally on the edge of the river, so there is nothing between you and the water. It is beautiful and definitely my new favorite route for running!