published on October 27th, 2019
I'm writing this post at the tail end of my first "ultralearning" week where I'm trying to learn data science.
Overall, the week went pretty well. I spent about 14 hours on the project - a bit less than what I had intended, but there were some unforseen things which made it impossible to meet my full committment. I've worked pretty concentratedly, being aware of the limitied time and setting timer has worked in the first week - so I'll keep doing it.
I've watched three lectures of the fast.ai machine learning introduction course and have practiced the question book method and open recall which worked really well to remember what was in the lectures.
I also submitted to three Kaggle challenges as planed. I really stumbled through the kaggle submissions and as a result, the submissions were really low quality - at least it should not be too hard to improve them. My biggest challenge is that I don't know at all what to do to improve my predictions. I fumbled around with some random forest parameters to improve the results on the validation set, but this was really just try and error. I'm also stumbling more than I thought working with numpy and pandas, getting a routine when working on the kaggle challenges will make this work faster.
I'm happy with my progress, given that I've only worked on this for approximately 14 hours, which is less than two days of full time work. There is of course a lot to be done.
In terms of tangible results, I have the following to show
Titanic: categorization accuracy 0.77033, position 7499 / 12930, 58th percentile.
Rossmann: RMSPE 0.20313, position 2783 / 3303, 85th percentile.
House Prices: RMSLE of 0.15678, position 3229 / 4854, 67th percentile
I've watched the lectures and practiced retrieval via the question book method and open recall. This worked well.
I haven't written anything about learned concepts yet and haven't created any Anki flash cards.
Here are some very brief notes I took on the days of the week:
Here are my goals for the comming second week:
To a successful second week of ultralearning!