Batch Jobs

Miscellaneous: Special Batch Jobs

  1. Batch Jobs:

    Creating and running .xpl files.

  2. Special Batch Jobs:

    For verification and comparisons.

  3. Scripting:

    How to create automatic scripts.

Contents

Create a Batch PAPPI History File
Processing the PAPPI History File

Create a Batch PAPPI History File

  1. Create an .xpl (Example) file. Name it batch.xpl in the same directory as the PAPPI executable. 

    "A Course in GB Syntax"
BY LASNIK AND URIAGEREKA


THE THETA-GRID AND X-BAR THEORY
 [1:4a]	 John slept
 [1:4b]	 John hit Bill
 [1:8a]	*John slept Bill
 [1:8b]	*John hit
...
BINDING CONDITIONS AT LF?
 [6:64a]	*He[i] likes everyone that John[i] knows
 [6:64b]	 Everyone that John[i] knows he[i] likes	% Original: At
LF. Problematic: Argues for adjunction to IP
 [6:65a]	 John likes every picture of himself
 [6:65b]	 Every picture of himself[i] John[i] likes	% Original: At
LF. Problematic: ditto.

    Here, we use a copy of l&u.xpl.

  2. Start up a fresh PAPPI session. In the input window, type META-b. (This command is not available through the usual menu system.)

  3. PAPPI will run through the list of sentences in batch.xpl in a special mode. The history will contain additional information (compared to using the Run Batch button from the Run menu). In particular, the run time and the various counts for each parser operation are also reported. For example:

    Also, the start and completion times of the run are displayed in the history and the terminal window in which PAPPI was started. For example: 

    CPapp
Please wait. Loading defaults...
Start: Thu Aug  6 12:07:24 1998

End:   Thu Aug  6 12:15:31 1998

  4. Save the history using the Save History.. button in the History menu:

    Choose an appropriate .phf (PAPPI History File) file name.

Processing the PAPPI History File

Various awk scripts can be used to extract useful information from the saved PAPPI history file. Here is a list of useful C-shell aliases: 

more ~/.pappi
alias times grep "'Time taken:'" '\!*' "|" awk "'{"s += '$4'} "END {print NR, s, s/NR}'"
alias level1 "egrep -e '^content: (Parsing|LF|Parse blocked by)' \!* > /tmp/s"
alias level2 "egrep -e '^content: (Parsing|LF|Theta C|Condition A|Control|FI:|Parse PF|Assign S|Expletive L)' \!* > /tmp/s2"
alias levels "level1 \!* ; level2 \!*"
alias stimes grep "'Time taken:'" '\!*' "|" awk "'{"print '$4'} "' | sort -n"
alias reflevels "level1 \!* ; level2 \!* ; mv /tmp/s /tmp/t ; mv /tmp/s2 /tmp/t2"
alias compare1 "diff /tmp/s /tmp/t"
alias compare2 "diff /tmp/s2 /tmp/t2"
alias compare2c "diff -c /tmp/s2 /tmp/t2"

times
Prints the number of sentences processed, the total time in seconds, and the average time taken per sentence. For example: 

times new.phf
335 351.89 1.05042

indicates that 335 sentences were processed in new.phf. Total CPU time for the batch job was just under 6 minutes and the average time per sentence was just over a second.

stimes
Returns a sorted list of individual sentence times: 

stimes new.phf
0.01
0.01
0.03
0.04
...
8.93
10.97
29.54
115.02

reflevels, levels, compare1 and compare2
Used to verify and compare PAPPI builds before and after significant changes.
  1. Run the old build and use reflevels to extract and store away the results of the run.
  2. Next, run the new build and use levels to extract the results.
  3. compare1 will do a first level diff on the results with respect to grammaticality.
  4. compare2 performs a more in-depth comparision to also include the behaviour of each parser operation.
For example: 

(Assume reference build has been run and the history saved in ref.phf.)
reflevels ref.phf
(Assume new build has been run and the history saved in new.phf.)
levels new.phf
compare1
(No output. All the sentences have the same number of parses.)
compare2
214c214
< content: Theta Criterion                  6     2 D-structure Theta Condition      2     2 Case Filter                     52     6 Case Condition on ECs            6     6 
---
> content: Theta Criterion                  2     2 D-structure Theta Condition      2     2 Case Filter                     10     2 Case Condition on ECs            2     2 
(Here, the new implementation differs from the old in detail. For
example, the Theta Criterion was called 6 times instead of
twice.)

compare2c
compare2c is just like compare2 with the -c option. Using three lines of context, may provide a little more information: 

compare2c
*** /tmp/s2     Thu Aug  6 13:30:28 1998
--- /tmp/t2     Thu Aug  6 13:30:07 1998
***************
*** 211,233 ****
  content: Parsing:  [1:30c]     I wanted the bus to arrive on time 
  content: LF (1):
  content: LF (2):
! content: Theta Criterion                  6     2 D-structure Theta Condition      2     2 Case Filter                     52     6 Case Condition on ECs            6     6

Here, we see that the difference occurs when parsing sentence [1:30c].

Batch Jobs