search

daxiongshu / Coupon-Purchase-Prediction

2 stars 2 forks source link

MAP10 #4

Open daxiongshu opened 9 years ago

daxiongshu commented 9 years ago

Hi, in your code, CV/average_precision.py line 28 if p in actual and p not in predicted[:i]:

Isn't that "p not in predicted[:i]" always true?

Thank you

SudalaiRajkumar commented 9 years ago

Hi Carl, I just took the code from Ben Hamner's repository.

It is just an extra check to ensure that same prediction is not repeated twice. If it is repeated twice and also if this condition is not checked, it will then give a high precision. So it is better to have this in the code.

daxiongshu commented 9 years ago

I see, thank you for the explanation. Sorry, I wasted our slots again.

I'll move to a more cv-based approach. : )

SudalaiRajkumar commented 9 years ago

Carl, no worries. I am happy to see the improvement. Thanks for putting in your efforts to take us to third place. Moreover, anyhow am not using the submissions. So it is good that you are using it :) I will let you know in case if I need one or two submissions. Please feel free to use all the submissions till then.

I got stuck with some office work and so couldn't do much till now. I will also start contributing from now. I have few more ideas to try and will let you know once I try them, My ideas are as follows:

  1. Since we see a good improvement using the individual customer id, I will use customer id itself as one of the feature and so it will also get a weightage in prediction.
  2. Use the location based features as features
  3. Use the search history of the customer as features. I am sure this will definitely add a lot of value for our models.

Thank you again.

SudalaiRajkumar commented 9 years ago

Carl, is it fine if I take one submission today. I wont be needing more than one. Thank you.

On Tuesday, September 8, 2015, carl notifications@github.com wrote:

I see, thank you for the explanation. Sorry, I wasted our slots again.

I'll move to a more cv-based approach. : )

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-138611020 .

daxiongshu commented 9 years ago

Sure! Go ahead please : )

SudalaiRajkumar commented 9 years ago

Carl, Adding just the cust_id did not help. I will try adding more new features and once that is done, I will start doing bagging. The results vary a lot from 0.009 to 0.010 !!

By the way, do you use WhatsApp?

daxiongshu commented 9 years ago

Hi, thank you for your work. I'll start building on CV for my models. What is your cv score of sub10.csv for the last week?

SudalaiRajkumar commented 9 years ago

Thank you Carl. Basically, at that time, I didn't get a CV.. I will compute the CV score and let you know the value..

Also I am trying to create few more variables based on visits. Most probably over the weekend, I will complete both. Will let you know once am done. Thank you.

Thanks, Sudalai

On Friday, September 11, 2015, carl notifications@github.com wrote:

Hi, thank you for your work. I'll start building on CV for my models. What is your cv score of sub10.csv for the last week?

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-139563277 .

daxiongshu commented 9 years ago

Thank you. Don't worry about it. I'll test it anyway. So is the last week 2012, 06, 17 ~ 2012, 06, 23 as in sub10 folder?

By the way, I'd like to skype you (video talk) this weekend. I can do it tonight, your morning tomorrow or any time you are available.

Thank you.

daxiongshu commented 9 years ago

I can reproduce sub10 now :D

SudalaiRajkumar commented 9 years ago

Yes Carl. The last week is from 17-3 to 23-3 as you mentioned.

Nice.. Good that we are able to reproduce sub10.CSV now. I feel more relieved now :D

I am traveling right now and so might reach late in the morning. I will drop a mail once I reach so that if you are awake, we can Skype. Else we can do over the weekend when you are free. Thanks again.

Btw, just saw the heated discussion in dato forums! How do people get such a high value.? If they get it, I think you can easily get it :) all the best for that as well.

On Friday, September 11, 2015, carl notifications@github.com wrote:

I can reproduce sub10 now :D

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-139582075 .

SudalaiRajkumar commented 9 years ago

Do you use WhatsApp??

On Friday, September 11, 2015, Sudalai Rajkumar SRK ssraj.rox@gmail.com wrote:

Yes Carl. The last week is from 17-3 to 23-3 as you mentioned.

Nice.. Good that we are able to reproduce sub10.CSV now. I feel more relieved now :D

I am traveling right now and so might reach late in the morning. I will drop a mail once I reach so that if you are awake, we can Skype. Else we can do over the weekend when you are free. Thanks again.

Btw, just saw the heated discussion in dato forums! How do people get such a high value.? If they get it, I think you can easily get it :) all the best for that as well.

On Friday, September 11, 2015, carl <notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com');> wrote:

I can reproduce sub10 now :D

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-139582075 .

daxiongshu commented 9 years ago

Hi, I just downloaded it. How can I find you?

For dato, I think they found a golden feature, or a leak. Maybe some javascript stands for sponsored page in the html. I don't know : )

SudalaiRajkumar commented 9 years ago

My number is +91 995 233 2232.

Please send your mobile number. I will contact you.

On Friday, September 11, 2015, carl notifications@github.com wrote:

Hi, I just downloaded it. How can I find you?

For dato, I think they found a golden feature, or a leak. Maybe some javascript stands for sponsored page in the html. I don't know : )

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-139585528 .

daxiongshu commented 9 years ago

Hi, my number is +1 4123203304

daxiongshu commented 9 years ago

Hi, could you send me the test_predictions_xgb.csv of the new 0.104 model, I'd like to run a bagging with it. Thank you!

SudalaiRajkumar commented 9 years ago

Carl,

I have uploaded the file in the shared gdrive which you have shared sub_0.0104.7z is the file name

Thanks, Sudalai

On Tue, Sep 15, 2015 at 1:44 AM, carl notifications@github.com wrote:

Hi, could you send me the test_predictions_xgb.csv of the new 0.104 model, I'd like to run a bagging with it. Thank you!

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-140193362 .

SudalaiRajkumar commented 9 years ago

Also I have uploaded the predictions of my second submission which I made today sub_0.0100.7z in the shared folder

On Tue, Sep 15, 2015 at 2:17 AM, Sudalai Rajkumar SRK ssraj.rox@gmail.com wrote:

Carl,

I have uploaded the file in the shared gdrive which you have shared sub_0.0104.7z is the file name

Thanks, Sudalai

On Tue, Sep 15, 2015 at 1:44 AM, carl notifications@github.com wrote:

Hi, could you send me the test_predictions_xgb.csv of the new 0.104 model, I'd like to run a bagging with it. Thank you!

— Reply to this email directly or view it on GitHub https://github.com/daxiongshu/Coupon-Purchase-Prediction/issues/4#issuecomment-140193362 .

daxiongshu commented 9 years ago

Thank you very much. Could you upload the new rf prob file please? Thank you.