Is there a way i can score my results offline without uploading to check?
How about dividing your training data into the train and validation sets? If your validation set resembles the test set, you should get useful feedback how your model is doing.
i already split data into train and validation data set. The mae score is diffrent from the online by zindi
You cannot expect to get the same results from validation and test sets. However, if your score on validation set is very different than the score you get from the test set, then most likely something is wrong with the validation set. Maybe too small, maybe not representative. Try to construct the validation set such as it resembles the test set.
You might want to Develop your own evaluation strategy from the given data.... e.g. using crossvalidation approaches in python.
Then build a "naive" model like what @Pawel_Morawiecki shared below as your baseline and benchmark against that. According to Pawel his model scored around 3.93 of the metric (mean absolute error).
Hopefully your offline evaluation strategy gives you a score close to that(3.93).....Your task will now be to create a model that beats this baseline in your offline evaluation. If your model is better than this baseline.... your submission should score you less than 3.93 mae. (!mind overfitting hazzard)
When you think of the task in a real world scenario.... the testset is data you will never know beforehand because it is in the future.
you must be online for you to upload your stuff. otherwise you cannot