# 7th Place Solution

Approach
The best part of the competition for me was feature engineering. In the end  I used 51 features, truncated down from 150.
By far  strongest features were base distributions of each Assessment. However, they were used by everyone. Individual features, given small data, were not so important, but they decided winners I think.
Final model was an ensemble: 0.3 LGB, 0.3 CATB, 0.4 NN
20 fold-bagging for all models, for NN additionally averaging across 3 seeds
One “trick” – use assessments from test set “blindly” (as we did not see this data) as samples in training. Data was scarce – so I was looking for every way to increase the number of training samples. Especially that you add data exactly for the children that are in the private LB.
Results
Truncated CV: 0.575
Private LB: 0.559
Public LB: 0.559
Final Remarks
Congrats to winners - looking forward to your solutions
Thanks to the organizers for the competition with event data – love those 😊 It unleashes your creativity in feature engineering
Validation setup
Truncate - select one assessment randomly for every child to reflect test set structure
Update 1: Features
Motivation:
I was impacted by the ideas presented in this paper by Francois Chollet:
[On the measure of intelligence](https://arxiv.org/abs/1911.01547)
. There are tons of interesting and powerful thoughts there. I was mostly stimulated by a discussion on how to measure intelligence: 
- A/ by overall-skill-level 
- B/ by skill-acquisition-tempo
In our case, we are measured by A/, which can be broken into two drivers:
- experience, i.e. how much time/effort the child has spent on various actitivities in the game. This formed my first group of features
- accuracy - how accurate was this child in her journey. This formed my second group of features.
However, skill-acquisition-tempo is a very interesting way to capture how quickly children are learning (features like minutes per level, events per level, etc.). This formed by 3rd group of features
I love competitions with manual feature engineering. Combination human+machine wins, which represents my view on how AI will impact the world.
Update 2: Feature selection
Calculated cv score after dropping a feature - did this individually for all ~150 features
Dropped all features which brought an improvement of less than 0.0001 on QWK score - I treat them as noise. Found ~100 noise features in this way.
Recalcuated  CV once more to see that overall score improved slightly after removing 100 noise features