pddasig / machine-learning-competition-2020 Goto Github PK

Hello,

Where can we do a submission and how? Is there a platform for this?

thank you

Please open a discussion here.

Dear PDDA Committee,

We have a question regarding the features to be used in the model. Must we use all seven provided logs? Or could we reduce the number of features?

Best regards,
Rock Abusers

After spending some time inspecting the train.csv data set, I am curious if it is intentional that the DTC & DTS (target response variables) have gaps? I understand the goal is to predict these curves when they are missing, but it seems that having them missing in the training dataset is suboptimal?

Here is the data contained in train.csv. Other than replacing the -999 values with NaN's, no editing has been performed:

Referring to the example notebook, it appears that there are no missing values in either the DTC or DTS:

Additionally, the PE curve seems to effectively end after approximately index 20,000. When the rest of the shallower data is removed due to the missing segments in DTC & DTS, it seems to reduce this down to using CAL, GR, CNC, HRD, HRM, & ZDEN.

My primary concern is if this is intentional, then it seems to remove a substantial amount of samples for training a model given this is based on a single well. Could you please comment on this?

Thanks!