Trouble shooting the UDpipe Model

The part of sentence (POS) tagging performed in PID.POS leverage’s the pretrained udpipe model framework. This model is not included in PID.POS due to its size, and hence has to be downloaded when the package is first used. For most users this process is seamless, but if users are behind a firewall or have limited permission to write to the hard-drive the automated model download may fail.

Here we provide brief advice on how to diagnose and fix common problems.

Step 1

When first installed and loaded PID.POS will not have the udpipe model. The first time data_frame_report is run, an attempt is made to download the model. Hence, it is recommended to start by evaluating the data_frame_report function with the example data (the_one_in_massapequa):

library(pid.pos)

report <- data_frame_report(the_one_in_massapequa)
browse_udpipe_repo()
browse_model_location()

If the model is not downloaded, an error is thrown signposting the functions browse_model_location and browse_udpipe_repo.

Step 2

Assuming the data_frame_report throws an error, the next step is to check if the model is present. This can be done by running browse_model_location() and checking for a file named ‘english-ewt-ud-2.5-191206.udpipe’ or similar. If the user does not have permission to write to this location, the model will not be downloaded and browse_model_location may throw its own error. If this happens, there are two approaches we suggest:

1. Contact your system administrator and have them either grant permissions or add the udpipe model to this location e.g. by running the code on your behalf.
2. Change the default model location using set_model_folder('new/path').

The former of these is the more reliable, and will allow PID.POS to operate as intended in future sessions. The latter is a temporary fix, and will need to be repeated each time the package is loaded.

Step 3

Assuming the user has permission to read and write from the model location, the next possibility is that the user is behind a firewall. In this case, there are two steps to consider:

1. Contact your IT team and ask them to allow downloads from github.com
2. Navigate to the udpipe repository via browse_udpipe_repo() and download the model manually. It will then need to be deposited in the folder opened by browse_model_location().

Step 4

If further problems persist please raise an issue on the PID.POS github page. We will be striving to expand this document as new problems/ solutions present.