search

seg / 2016-ml-contest

Machine learning contest - October 2016 TLE
Apache License 2.0
187 stars 269 forks source link

Test Submission #101

Closed davidgtang closed 7 years ago

davidgtang commented 7 years ago

Hi Matt,

Our team (Carlos Fuerte) is trying to submit to the TLE contest. Hope we're submitting it the right way. Thanks!

kwinkunks commented 7 years ago

Hello David! Thank you so much for the entry — I love seeing new teams trying the contest. Everything worked perfectly, and thanks for saving your prediction out into a CSV. Your entry scored 0.480, but I noticed you had not re-trained your model with all the data, but were still using the training subset. So I did this at the end of your notebook:

clfNN = MLPClassifier(solver='lbfgs', alpha=.015, hidden_layer_sizes=sizes, random_state=9)
clfNN.fit(scaled_features, correct_facies_labels)
y_pred = clfNN.predict(scaled_validation)
np.save('y_pred_Matt.npy', y_pred)

That random state does a bit better than other values I tried, so there's some stochastic 'wash' in there. You should be able to reproduce the entry I actually scored with the code above — it gets 0.561, which is a really good first effort. Nice! I bet with some of the feature engineering others are experimenting with you can improve on that.