skrutable 2.3.0__tar.gz → 2.5.0__tar.gz

{skrutable-2.3.0 → skrutable-2.5.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: skrutable
-Version: 2.3.0
+Version: 2.5.0
 Summary: skrutable library for working with Sanskrit text
 Home-page: https://github.com/tylergneill/skrutable
 Author: Tyler Neill

skrutable-2.5.0/src/skrutable/__init__.py

@@ -0,0 +1 @@
+__version__ = "2.5.0"

{skrutable-2.3.0 → skrutable-2.5.0}/src/skrutable/config.json

@@ -29,8 +29,9 @@
 	"upajāti, non-triṣṭubh, perfect" : 4.5,
 	"upajāti, triṣṭubh-jagatī-saṃkara, perfect" : 4,
 	"upajāti, non-triṣṭubh, imperfect" : 3,
-	"jāti, perfect" : 8,
-	"jāti, imperfect" : 3,
+	"jāti, perfect" : 9,
+	"jāti, imperfect" : 5,
+	"jāti, likely" : 3,
 	"none found" : 1
 	},
 "preserve_punctuation_default" : true,

{skrutable-2.3.0 → skrutable-2.5.0}/src/skrutable/manual.md

@@ -130,6 +130,31 @@ Key terms:
 * *jāti*: four quarters with set patterns of total moraic length
 
 
+# scan timing profiling
+
+`skrutable` includes a built-in profiling system for measuring meter identification performance across a corpus. It is disabled by default (`utils._DEBUG_TIMING = False`) and has no runtime cost unless explicitly enabled.
+
+To use it, set `_DEBUG_TIMING` before importing `MeterIdentifier`:
+
+```python
+import skrutable.utils as _utils
+_utils._DEBUG_TIMING = True
+
+from skrutable.meter_identification import MeterIdentifier, flush_profiling_report
+
+MI = MeterIdentifier()
+for verse in my_verses:
+    MI.identify_meter(verse, resplit_option="resplit_lite", resplit_keep_midpoint=True, from_scheme="IAST")  # e.g.
+
+flush_profiling_report()  # prints table to stderr, resets counters
+# flush_profiling_report(write_file=True)  # also writes profiling_debug.txt
+```
+
+The table breaks down wall-clock time per meter category with columns for each scan sub-phase (`clean`, `transl`, `syl`, `wts`, `mor+g`) and each identification type (`anuṣṭ`, `samav`, `jāti`, etc.), plus perfect/imperfect verse counts per category.
+
+If using the front end, `make launch-profiling` (or `./launch.sh --scan-profiling`) enables profiling for the server process without touching any source files.
+
+
 # sandhi and compound splitting
 
 `skrutable` provides a wrapper for applying pre-trained splitting models via separate online servers ([my own splitter_server for the 2018 model](https://2018emnlp-sanskrit-splitter-server.duckdns.org/) and https://dharmamitra.org). A working internet connection is required for this functionality. The wrapper preserves original sentence length and punctuation, and it also helps utilize the Dharmamitra ByT5-Sanskrit model's ability to distinguish compounds from inter-word breaks.