Hi,
I worked on last year dataset and it was normal dataset, I used the best model I used on 2018 data set on 2019 data set, it return a very strange result:
the result was, after training, the validation accuracy is about 89.xx but the classification report from sklearn.metrics was very bad.
this report contain recall, precision, f1-score, all are under 0.50.
i use the same model again on the 2018 data set and the result was good.
question : is there any changes on the 2019 data set from the 2018 data set.
thank you
Hi @alla1g2f0,
I’m not totally clear on your question, but let me clarify some things about the 2018 vs 2019 data:
- The training data from 2019 is a strict superset of the 2018 training data. More specifically, the 2019 training data includes images from the following sources:
- “BCN_20000” - this is brand new training data
- “HAM10000” - this is the 2018 training data for Task 3 (classification)
- “MSK” - this is previously-released training data from 2016, 2017, and 2018 Task 1 and 2
- The test data from 2019 has not yet been released, but it will include many new images that are not part of the 2018 test data.
I hope this helps. Please let me know if you have any other follow-up questions.
1 Like