Update description.md

description.md

@@ -1,14 +1,3 @@
 # Description of submission
 
-This example demonstrates a problem that is likely to be quite common. There are many cases when some category of a categorical variable is present in the holdout data, but absent in the training data. If you are creating dummy variables at the preprocessing stage (or using one-hot encoding, etc.), applying the same clean_df function to the training and to the holdout data will lead to different results - to the datasets with different sets of variables. This will lead to the failed run on the fake data and on the holdout data. 
-
-
-In this example we use LASSO regression. It requires all categorical variables to be turned into dummies. 
-We include a variable cf20m180 (satisfaction with relationships with the current partner) in a model. 
-If we select only cases in the training data for which the outcome is available, there are no people who answered 0 and 1 (very dissatisfied). This is why there will be two columns missing 2 (for cf20m180==0 and cf20m180==1)
-These categories however appear in the full train set - and are likely to appear in the holdout data, in which case it will introduce errors.
-
-In this example we create additional factor levels to deal with this problem.
-
-
-UPD: plus test adding external data using fake external data
+Example submission - with correct model.rds (in previous version one variable was missing, gender_bg)