#mitsui-commodity-prediction-challenge

1 messages · Page 1 of 1 (latest)

unkempt onyx
#

anyone else's baseline submission taking forever?

simple prairie
#

Looking for serious people to colab with for this contest

#

to get top 20 and work 2-3 hour per day or work hard

leaden mountain
#

Hey @simple prairie i am interested in joining you.

hallow mural
#

Hey @simple prairie I am interested to join you

simple prairie
#

dm

weary gulch
#

Hey @simple prairie I am interested in joining you

vivid holly
#

Looking to team up for this comp. Would be great to work on a shared repo and maximise learnings and exploring interesting techniques rather than chase leaderboard. Especially anyone who is based in London and is willing to do in person coding sessions

broken flare
#

Hey kagglers,

If you're diving into the this comp - check out my latest discussion post on why WarpGBM (my co-created GPU-accelerated GBDT lib) could be a game-changer—especially its invariant learning for handling market regime shifts and lagged time-series data.

It offers ultra-fast training, robust generalization, and easy integration. Perfect for boosting your models in volatile commodity forecasting!

Discussion here: https://www.kaggle.com/competitions/mitsui-commodity-prediction-challenge/discussion/593904

Repo: https://github.com/jefferythewind/warpgbm

MITSUI&CO. Commodity Prediction Challenge

Develop a robust model for accurate and stable prediction of commodity prices

GitHub
GitHub - jefferythewind/warpgbm: WarpGBM: High-Speed Gradient Boosting

WarpGBM: High-Speed Gradient Boosting. Contribute to jefferythewind/warpgbm development by creating an account on GitHub.

past inlet
broken flare
# past inlet According to you which model works best for this competition?

I'm not sure yet will have to run experiments but I think invariant learning from warpgbm should outperform other gbms in terms on generalization but ideally an era wise ensemble of different models should be a good bet

In numer.ai competition which is similar warpgbm does pretty decently in preliminary testing

The primary issue with this comp is the lack of public test so we have to really on generalization solely based of CV which makes honest experimentation the most important aspect

past inlet
broken flare
mortal elbow
#

Hello everyone, looking to team up with someone based in Cali or elsewhere in the US. I can contribute several hours most days of week after 4 pm PST. Please let me know!

mint valve
jovial pulsar
drowsy atlas
#

Currently studying the impact of technical indicators and thier effectiveness in building stock predicting models. Looking for team mates to help study other critical areas that will help our model and what approach to use. If interested, kindly send a DM

red umbra
#

Good afternoon, currently failing after receiving "notebook inference server errors" (I've never gotten past this) after submitting my submission.parquet. I've got a good model team with LightGBM and XGBoost, thinking about adding a training model to minimalise RMSE. I know the competition is judged on the Spearman rank correlation...

mortal elbow
#

Hello, currently trying to do some feature engineering, but I think the current data does not have enough "info" to actually derive something useful. I imagine it's fine to make API requests for other datasets or do we have to use the included data only? The predictions are accurate only to a certain degree, but if I can obtain more data from other sources, I would make things a lot better

red umbra
#

@mortal elbow same boat I'm in. This is a very tough challenge with the quirky API.

jovial pulsar
red umbra
#

yeah good luck with that shit, I've got a boilerplate working, and the advanced featureset is getting tough to get my inference speed down to where I'm not flunking

#

vectorized prediction and batch processing debugging today, how is everyone else doing?

red umbra
#

GTFO scammer

red umbra
#

yay! finally got on the LB at #333!!

vivid holly
#

Looking for last member for Mitsui comp, looking for someone who has past Kaggle experience and prioritises learning over chasing leaderboard scores.

red umbra
#

oh look more trash made its way in here!

formal olive
#

In this notebook,his trainning data included test data,right?

#

he didnt split it

red umbra
#

Looks like it was structured to be a complete end-to-end workflow, so yeah.

#

Trained the prophet model using only the training data and fitted a model to predict values for the test set dates.

#

look at the finals cells - they're dedicated to putting the predictions into a submission.

#

I tried this approach myself and got pretty shit results.

red umbra
#

finally got that score up to the .7 range!! How's everyone doing?

red umbra
#

GTFO scammer

pseudo karma
#

Hey everyone, I just arrived, and Mitsui Commodity Prediction is the first competition I'm participating in.

#

Is this a good score?

{
evaluation_method": "mean_target_correlation / std_target_correlation"
"enhanced score": 0.8383663342668812,
"baseline_score": -0.001835422346098309,
"improvement": 0.8402017566129796,
"mean_target_correlation": 0.0701314224109782,
"std_target_correlation": 0.0836524792855684,
"num_valid_targets": 424,
"positive_targets": 328,
"strong_targets": 157,
"excellent_targets": 27,
}
#

As it's my first competition and I’ve never used the Kaggle notebook interface, I would like someone to help me with the submission process.

#

:(

red umbra
#

anyone else having difficulty adding the mitsui input set

#

I go to add the input.. and get "error: internal error"

#

yeah the data source won't even open on an empty notebook. Anyone else?

red umbra
#

Hi Daniel, please GTFO and don't come back.

red umbra
pseudo karma
#

Got 0.34 of public score is this good?

scenic garnet
hardy dirge
#

Quick question, assuming there's a date t, targets A,B,C and D. Is it allowed to predict A at t using B,C and D?

vernal tendon
vernal ivy
#

Hello, This is Chirag. I hope all is good. Noob here.. trying to level up via this competition

serene kelp
#

Hi! Anyone still up for teaming? Want to learn and progress together. Currently 434 on the leaderboard. DM me if interested. Score 3.030

red umbra
#

@serene kelp I'm right there with you in the same scoring area, good luck to you!

red umbra
#

Get this bullshit out of here.

civic elk
#

hey i am new to these kaggle challenges. Can anyone please help me get an idea what the submission output should look like. Is it 1 file, date x targets long or is it 4 files each of them being dates x targets_lag(x) long where x is 1-4.

strong storm
#

hey, i am new to kaggle and a noob overall. I've tried to run a baseline submission. I submitted the demo submission and it worked fine. Then I changed the definition of prediction, replacing the i / 1000 part with 0 to get a baseline. But I keep getting submission scoring error. I've also tried 0.0 and 0*(i/1000), same error.

fiery tide
#

Hello, I want help with something cause I am kinda new submitting to this type of comp
I have been troubleshooting a Submission Scoring Error all day and now I official got Kaggle Error during submission, with the exact same pattern as demo submission and the resulting parquet file 1-1 with the demo submission in terms of schema (134 rows, 425 cols), date_id as int64 and the others as floats (essentially my local gateway works fine but for some reason kaggle hates me). If someone more experience wants to help me I can provide my nb in dms.. and I would pretty much appreciate it

fiery tide
fiery tide
strong storm
clever latch
#

any1 wanna team up with me for this competition

brisk mesa
#

Hey Team,
After submitting notebook seeing
Notebook inference server never started. Notebook is running for a long time. Is it because we need to call inference server module in first 15 min? If so do I simply call without predict or how should I approach this? This is my first time with inference server. Would really appreciate your inputs in this.

rapid hamlet
#

i've fixed this problem by adding a random term becauseapparently I was diving by 0 variance. I think I'm missing something crucial. Can someone explain why I'd be dividing by the variance of the returns of the targets?

vernal ivy
#

any one doing local validation and able to match submission score?

hoary wraith
#

Yo what’s the difference between entrants and participants in the competition?

hoary wraith
#

also how do you download the data lol

#

Btw just to be clear Im not the daniel people were getting mad at earlier (I literally just joined like 5 days ago lol)

sick elk
#

what was about today ? event

#

some point you can give me

broken flare
#

Where was this event?

frigid basalt
#

Hi

sour summit
#

!rank

sullen sage
#

Hi

cinder citrus
#

Hey, anybody willing to discuss about industrial project ? I need help 😫

strong storm
#

Anyone's notebook crash/booted out the leaderboard after the final rerun?