but it is really impersive, I couldn't exploit the data as much as that

Hmm, true tho

what method's u guys using to exploit the data beyond the score of 30 rmse?

I have an ensemble model, of lstm, arima and sarima

Since you need to explain how the features are contributing to the predictions, I think an ensemble is out of the question for this competition.

Ooh I will have to change the model used then.

@yisakberhanu how did you crack less than 20 let's start there 😂

@Koleshjr The moment he started getting good scores, he vanished from the forum.

@yanteixeira he should start by example by sharing how he cracked less than 20😂😂anyways I just found out that someone has scored 16 with no features, Maybe the rest of us are complicating things.

I have no idea how someone could get a good score without using features. The post-competition period will be extremely resourceful in terms of learning. I’m very curious to see people's solutions.

- Quick hint: You can reach a score of 24 without a model. Once you achieve this score, you can build your model upon it.

- Quick hint 2: Prophet is probably a bad choice for this competition.

Yan, I'm having trouble submitting, i used a traditional ML model and I'm trying to make predictions for the dates in the sample submission data. After extracting the dates and trying to predict the clicks i got a feature mismatch error. I'm new to time series, please how can i make predictions for these dates?

@AI_Maven It's impossible to help without seeing the code. it could be anything. I suggest you read the discussion section. I'm sure you will find useful information there.

@Jaw22 I propose the following: share a sample code that achieves an RMSE of approximately 13 to 13.50 in the AirQo African Air Quality Prediction Challenge, and I will share a sample code that reaches 24 in this competition. Since exchanging information privately is against the rules, both solutions should be made public.

What I'm saying is the shape of my final training data(the train.csv file for this competition) is (67968,14). But if you extract the dates from the submission file, the shape of the dataframe will be (370,1). Originally for you to make predictions for the test data, the features should match that of training set. Now I can't do this because i don't know what the features of the future dates looks like. That's why I'm confused, how then can i make predictions with my model. How are guys able to make submission? This is the issue I'm facing currently 🥲

@yanteixeira You could challenge me to do the same and then you publicly share the code implementation that achieves 24. Inserts laughing emoji!

I think he was challenging all of us😂😂 we should open source a notebook that achieves 13.50 and he will give us the 24 notebook

@yanteixeira 24 without a model are you kidding me 😂 you mean I have wasted 100 subs and I can achieve 24 without a model?

hahaha

@AdeptSchneider22 @Koleshjr Yes, this deal involves all of you. If someone posts a notebook with a score between 13 and 13.50 on the Air Quality Challenge, I will post a notebook here with a score of 24

@Koleshjr Sorry, my friend... I think I'm feeling the same about the other challenge. I'm just unable to score lower than 14, while you have achieved a good score of 13. Sometimes, a given competitor excel in one challenge but fail to find the best approach in another. That's life I guess..

that's true @yanteixeira . @AdeptSchneider22 are you up for the challenge or I take one for the team😂😂😂😂

Okay @yanteixeira I will take one for the team but let us make it fair. In AirQo the first person has 12.01 and you are asking for a nb that scores 13.50 thats an rmse difference of 1.50. In this you have 12.39 and you are saying you will open source a 24 notebook that is an rmse difference of 11.61/1.50 thats almost 10x , make it 5x and open source a 19 to 20 nb and we have a deal😂

I'm pretty sure that once you understand how to achieve a score of 24, you will quickly find your way to 20

Sure , but 10x is a lot , or a we make it 13.80 range for 24

I second this. This is fair play.

ok then, 13.80 range for 24.50 range

Nice we have a deal 🤝

@yanteixeira my end of the bargain is fulfilled, we are now waiting for yours

Thanks. It's a really good notebook, by the way. I'm at work right now. When I get home, I'll publish my notebook.

coool

Thank u @Koleshjr @yanteixeira @AdeptSchneider22.

Cool, @yanteixeira! I'll share a simple code to achieve 12... as well.

For AirQo African Air Quality Prediction Challenge

@mubarak127 thank's for the contribution!

@yanteixeira that is an unfair trade; let us just create a team and share our codes. than we not violating any rules...I am sure that model is over fitting!!!!

That's not unfair at all. In any case, I'm already wrapping up the notebook.

Do the maths yan...the pricing of airquality is more durable than the pricing of clicks.

😂😂😂I also wonder the same thing as well 24rmse without a model...crazy!!

@Jaw22 Okay, man, just don't use my code then

@Koleshjr @Ebiendele If you have any doubts, you can post them on my notebook post

cool!!, we'll be expecting👌👌