Can we train a model and test it with a data set having different no.of columns???
Maybe you can’t because Whichever variable use for train model need those variable only.
If you try doing that, you will get an error. This is because your model learned from a certain set of columns and expect a similar set in the test data to make predictions.
Your query is strictly true for the supervised learning tasks. However, we could always have more features and dimensions (that can be reduced to certain dimension or vector reshape ) in unsupervised problems. Note that Its different set of vector features. Not different no. of columns
In that case, select only the columns that you feel relevant. Another process you can follow is to fill up the missing column with some mean value for the feature and pass the input to your model.
Thank you all