.. _fwflag_turn_off_backoff_model:
==========================
--turn_off_backoff_model
==========================
Switch
======

--turn_off_backoff_model

Description
===========

Turns off the automatic "backoff" to a logistic/linear model when a prediction run has only a small number of features, forcing DLATK to use whatever :doc:`fwflag_model` specifies instead.

Argument and Default Value
==========================

No argument. Boolean switch, default: ``False`` (backoff model enabled).

Details
=======

Verified: ran cleanly as part of a full :doc:`fwflag_combo_test_regression` run alongside :doc:`fwflag_combined_models` and :doc:`fwflag_residualized_controls`, with no observed change in output structure — the backoff behavior only kicks in for feature counts too small to matter on typical corpora.

Other Switches
==============

Optional Switches:

* :doc:`fwflag_model`

Example Commands
================

Using the packaged ``dla_tutorial`` dataset (see :doc:`../../tutorials/tut_dla`):

.. code-block:: bash

	dlatkInterface.py -d dla_tutorial -t msgs -c user_id -f 'feat$1to3gram$msgs$user_id$0_05' \
	--outcome_table blog_outcomes --outcomes age --combo_test_regression --turn_off_backoff_model