16GB Intermediate solution: XGB Era Boosting

Tournament
41

I modified example_advanced_32GB.py as follows;

do ensembles

    training_data["ensemble_neutral_riskiest_50"] = sum(
        [training_data[pred_col] for pred_col in pred_cols if pred_col.endswith("neutral_riskiest_50")]).rank(
        pct=True)
    training_data["ensemble_not_neutral"] = sum(
        [training_data[pred_col] for pred_col in pred_cols if "neutral" not in pred_col]).rank(pct=True)
    training_data["ensemble_all"] = sum([training_data[pred_col] for pred_col in pred_cols]).rank(pct=True)
    training_data["preds_model_target_neutral_riskiest_50"] = sum([training_data[pred_col] for pred_col in pred_cols]).rank(pct=True)

    ensemble_cols.add("ensemble_neutral_riskiest_50")
    ensemble_cols.add("ensemble_not_neutral")
    ensemble_cols.add("ensemble_all")
    ensemble_cols.add("preds_model_target_neutral_riskiest_50")
42

much appreciated, resolved it. I was off chasing the error in utils thinking it was spitting it out when it was calling validation_metrics

43

@bigbertha which model do you use example_advanced_32GB.py or the intermediate?

44

I’m just getting caught up on all of this, I’ll try to get all of this updated in the next week or so. Holidays have me way behind. @mesomachukwu12 appreciate your help here. Thank you!

45

I have a different way of constructing my models from the way the example model is built. But I am using all available information on feature groupings (I really miss feature groups!)

2 Likes
46

@bigbertha, I will appreciate if you can share with me [email protected]

47

I’ve been giving this a lot of thought since yesterday’s TC announcement: I’m not sure I’m going to be able to reproduce the boruta output when the new classic data drops. The initial run took a little over a week and when the data 3x’s, that could obviously push out much further. This is being further complicated by my new signals pipeline which is turning into a compute black hole.

I just wanted to give you all a heads up and time to prepare for the next data drop. GL!

48

Thanks for the heads up. The challenge is good

1 Like
49

No worries your work is highly appreciated!

1 Like
50

Hey, can you explain why exactly you need to fill int8 with 2 instead of 0.5 in this case? Been trying to find the answer myself and can’ seem to sort it out

51

int8 = integers only, so instead of values 0,0.25,0.5,0.75,1.0, you have 0,1,2,3,4. So 2 is the middle/neutral value instead of 0.5.

52

Ahhhh yeah how embarrassing… here I was thinking it was something to do with the pandas fillna method.

Thanks!

53

I actually made the exact same mistake, don’t think anything of it.

2 Likes
54

Hey @objectscience, I’m not able to access the repo anymore. :frowning:

55

Almost sent you a message the other day, haven’t seen you on the forum in a minute. Hope all is well.

Things on this end have gotten a little dicey since the first of the year and I’ve largely had to pull away from the comp. I was having a really hard time staying on top of the things that needed addressed with the repo and with TC and the new data-set coming thought it was best to pull it, so people didn’t waste time trying to use something that is out of date.

I still have some of the output from the run though, can paste it here if it will help.