Frequently Asked Questions

Common questions and tips ‘n tricks that may not appear elsewhere in the documentation.

Plom Client software

Some windows/dialogs are awkward to resize on Gnome Desktop

You could try disabling “Attach Modal Dialogs” in the “Gnome Tweaks” app, but mostly this has been fixed recently by improving the way we use modal dialogs.

Identifying

How does the auto-identifier work?

Machine learning trained on the MNIST handwritten digit dataset. A linear assignment problem solver then matches the results against the class list. For this reason the class list csv file should not contain large numbers of additional students that are not in your class.

Student wrote a different paper; I cannot ID their new paper.

For example suppose Isla’s name was printed on paper 0120 but they wrote blank paper 1280 instead. Plom’s “prenaming” is now a “prediction” because of exactly this situation. Simply ID 1280 as normal. (If 0120 was scanned and is blank, you can note that it is blank within the Identifier)

I made a mistake identifying: how can I revert an ID?

The UnID operation is exposed in the beta Manager Tool -> ID Progress tab.

Test Preparation

For self-submitted work, should I start each question on a new page?

Yes. This is important because it makes it easier for markers to find the answers, especially if they need to use the rearrange pages dialog.

While preparing the test, we suggest writing something like “Please start a new page NOW” at the start of each question.

Marking

Why doesn’t Plom have fractional marks?

Because the main developers of Plom don’t use them. We feel that they tend to over-complicate marking schemes and that it is better to simply make your test out of a larger total rather than mess about with “1/2”s or “1/3”s. This also makes things more consistent when a large team is trying to mark using the same scheme / rubric. We admit that this is forcing our own ideas about marking onto others, however, after supervising the marking of many tests and exams, we feel that this is the right way to go (the absence of fractional marks, not the forcing people to do what we think)

Of course, if an energetic co-developer would like to implement fractional marks, then we won’t stop them.

Why do you have “+0” and “-0” as possible delta-marks?

(Current versions of Plom do not expose this feature.)

Mathematics use “epsilon” to represent small number, often in the context of limits. Some markers like to indicate to students via (say) “-0” that there is a small error in their work but it is too small to reduce their overall mark. Similarly some markers use “+0” to indicate that a small amount of progress has been made, but not enough to be worth a full point.

Scanning

Do I need to carefully pick out just the right pages when rescanning a bundle?

It depends. Suppose we have scanned “BundleA1” but some pages misfed, and do not appear in the resulting PDF file. In this case, its perfectly fine to rescan the whole bundle. Use a new bundle name, say “BundleA1-rescan”. Now process as usual. When uploading, plom-scan upload BundleA1-rescan will upload the new non-colliding pages and you’ll get a message about collisions (lots and lots of collisions!) Just ignore that and do not pass --collisions.

Server administration

My server sometimes has random disk I/O errors

Like this peewee.OperationalError: disk I/O error? Plom uses an SQLite database; it should not be run on NFS storage. Apparently “people” know this but we were just as “thrilled” as you probably are to discover it on a production server.

How can I get past SSL certificate errors?

E.g.,

SSLError(SSLCertVerificationError(1, '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: self-signed certificate (_ssl.c:997)'))

Or plom-create status might be showing you:

[!] insecure connection (self-signed or invalid SSL cert)

SSL is used to securely verify the identity of the server you are connecting too. If you’re only experimenting, you can bypass the checks by setting a special environment variable, e.g., in bash:

export PLOM_NO_SSL_VERIFY=1

The graphical client offers a mechanism to ignore SSL errors (at you and your users’ own risks).

For production servers, you’ll need to investigate how to setup SSL certificates.

Ok, how do I setup SSL certificates?

We are not experts on this topic, but LetsEncrypt is a good place to start.

How can I clone a server?

For example, how can I make another copy of a running server? One way is to copy the filesystem of the running server, then modify serverDetails.toml to change the port. Its also possible to make a new server from scratch that accepts scans intended for the old server. This is discussed next.

How do I reset my server to the pre-scanned state?

You need two things: the question_version_map.csv file which you can get with the command line tools: plom-create get-ver-map. This is important because Plom needs to know which versions to expect for which question. You can upload this to your new server using plom-create make-db --from-file old_qvmap.csv.

You will also need the verifiedSpec.toml which is harder to get: it can be extracted from the file system of your old server by copying specAndDatabase/verifiedSpec.toml.

There are two fields in verifiedSpec.toml that are probably not in your original spec file:

privateSeed = "0052084227513987"
publicCode = "302386"

Calling plom-create uploadspec verifiedSpec.toml to push this spec into the new server will (currently) populate those fields as-is, thus ensuring the server will be able to read in physical papers printed from the original server. In future, this might require more effort such as passing a --force.

If you do not have access to the file system of your old server, it should be possible to extract the publicCode from the QR codes of the printed pages. See the source code plom/tpv_utils.py for hints on how to do this. The privateSeed should not be necessary for this procedure.

One should be very carefully doing this sort of thing: the publicCode exists to make it difficult to accidentally upload the papers to the wrong server. This question shows you how to defeat that mechanism.

I messed up by double-scanning some papers and uploading and now I have collisions!

You can simply delete the collisions using the Manager tool.

I messed up by double-printing some papers, now I have collisions!

This is a serious problem which you should avoid getting into… If two students have written on (say) paper number 20, then you will get collisions at upload time.

Let’s suppose the scanned bundles are contiguous: that is, the two paper number 20s are not interleaved in the scanning process. For example, they are in two separate bundles. (If this is not so, have a cry and consider sorting and rescanning?)

At this point you have paper 20 “A” scanned into the system. Now upload paper 20 “B” (you’ll need to force with --collisions). Paper 20 “B” is sitting in Manager -> Collisions. (We know they are not interleaved).

Next: if you have only one version, you can simply use the Manager tool to assign the colliding pages to Test Pages of some unused test number (assuming you have spares; if not see below). That’s it!

I messed up by double-printing and I’m using multiple versions

This is similar to the above but we cannot simply push paper number 20 “B” into a spare unused paper slot (say 107). This is because paper number 107 will have different versions than 20.

A future version of Plom will allow you to instantiate arbitrary new rows of the database using the versions of paper number 20. Roughly: extract the relevant version numbers (use plom-create get-ver-map) for paper 20. Use those to make the brand new row, using a complete new test number: say 1020.

But for now, the workaround is:

1. Clone the server from the pre-scanning state (see question elsewhere), change the port.

2. Update the duplicated papers into the 2nd server.

3. Have your grading team grade on both (alternatively, have them do most of the grading on Server 1, then download the rubrics with plom-create and push those rubrics to Server 2.

Changing the spec later

Students have already written my assessment, can I split one of my questions up? Can I merge two questions?

Yes, although there is some work. Keep the old server up for now (“Server A”). Make a new server (“Server B”). Hack the spec to duplicate the public code from A to B (see instructions above for “resetting a server to the pre-named state”). Change the spec as you wish (with in the constraints of the papers you already have). If you have mono-versioned test, nothing else is required: upload the papers to Server B.

If you have a multiversioned test, its a bit harder:

1. extract the version map from Server A (plom-create get-ver-map -s plomA.example.com).

2. modify that version map for your new paper layout. For example, if you are splitting “Q5” (physically laid out as 5(a) on Page 11 and 5(b) on Page 12) into separate “Q5” and “Q6”, then they must both have the same version as the original Q5.

3. upload that version map to Server B when making the database. plom-create make-db --from-file my_new_vermap.csv -s plomB.example.com

4. Upload the papers to Server B.

I have already uploaded scans, can I split one of my questions up? Can I merge two questions?

Not easily. Currently we would suggest re-uploading to a new server following the instructions above.