Submission Error MagNet

MagNet: Model the Geomagnetic Field
1

I am getting this error and unable to resolve it. any ideas
File “/codeexecution/predict.py”, line 54, in aggregate_hourly
[“period”, feature_df.index.get_level_values(1).floor(“H”)]
IndexError: Too many levels: Index has only 1 level, not 2
I am using the Baseline architecture to inference, my model. I appreciate any help and thanks in advance.

2

You may want to use get_level_values(0) - solved the problem for me!

2 Likes
3

As for me, I had to get rid of period in the group by as well.

Do you think it means that the test data does not contain period like train_a, train_b or train_c?

1 Like
4

You should not assume the names in the period column, treat this as an arbitrary categorical value. @suyashk12’s solution above is the right way to get these out of the DataFrame’s index if you need the specific values.

2 Likes
5

Rk: they also forgot to put the numpy library in. Maybe they could correct the benchmark blog post in order to have a working model because as is I had to correct 3 typos to make their benchmark work which makes it loose its interest as an exemplary code.

6

What do you mean by this? It looks like numpy is imported in the first code cell.

Please post any typos here, we’d be happy to fix them.

7

I mean that from what I understood, we are meant to copy the last cell (cell 31) in a predict.py file that we include in the zip. I beleive that we don’t want to copy the other cells in predict.py.

In this last cell, I had to change three things

  • change get_level_values(1) to get_level_values(0)
  • remove the period group in the same line as the one where get_level_values is
  • Add import numpy as np

Then with these three corrections, the last cell is indeed good for a prediction.py code for submission. Apart from that the benchmark blog post is very nice so if you can fix (or explain why there are these) typos this would be great.

8

Are you sure you are running all the code cells in exactly the same order? As I mentioned import numpy as np is at the very top of the benchmark:
image

9

And as I mentioned I know that numpy is imported in the first cell but I think the last cell is special, this cell should be copied as is in a predict.py file and hence it should come with its own imports. If not why would you re-import panda and keras in the last cell ?

10

@Timothee thanks for flagging these issues! You’re right, numpy should be imported in predict.py.

I’ve also updated the aggregation line to:

agged = feature_df.groupby(
            [feature_df.index.floor("H")]
        ).agg(aggs)

You should see these changes today. Thanks again for flagging!