tpattern 0.1.0__tar.gz → 0.1.1__tar.gz

{tpattern-0.1.0/tpattern.egg-info → tpattern-0.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tpattern
-Version: 0.1.0
+Version: 0.1.1
 Summary: Open, auditable temporal & sequential pattern analysis (T-patterns) for sport and behaviour
 Author-email: Andrew Callaway <acallaway@bournemouth.ac.uk>
 License: MIT
@@ -75,7 +75,12 @@ model and clear reporting:
 
 ## Installation
 
-From source (until published on PyPI):
+```bash
+pip install tpattern
+```
+
+Add the guided-wizard extras with `pip install tpattern[gui]`. To work from source
+instead (for development or to run the reproduction scripts and shipped data):
 
 ```bash
 git clone https://github.com/ajcallaway/TPattern.git

{tpattern-0.1.0 → tpattern-0.1.1}/README.md

@@ -39,7 +39,12 @@ model and clear reporting:
 
 ## Installation
 
-From source (until published on PyPI):
+```bash
+pip install tpattern
+```
+
+Add the guided-wizard extras with `pip install tpattern[gui]`. To work from source
+instead (for development or to run the reproduction scripts and shipped data):
 
 ```bash
 git clone https://github.com/ajcallaway/TPattern.git

{tpattern-0.1.0 → tpattern-0.1.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "tpattern"
-version = "0.1.0"
+version = "0.1.1"
 description = "Open, auditable temporal & sequential pattern analysis (T-patterns) for sport and behaviour"
 readme = "README.md"
 requires-python = ">=3.9"

{tpattern-0.1.0 → tpattern-0.1.1}/tpattern/__init__.py

@@ -19,7 +19,7 @@ from .methods import methods_text
 from .guided import run_analysis, launch
 from .contrast import group_contrast, contrast_items
 
-__version__ = "0.1.0"
+__version__ = "0.1.1"
 
 __all__ = [
     "Observation", "read_observation", "read_sample", "read_table",

{tpattern-0.1.0 → tpattern-0.1.1}/tpattern/advisor.py

@@ -191,5 +191,51 @@ def recommend(observations, *, min_count: int = 15,
         evidence={"n_obs": n_obs},
     )
 
+    # --- Frequent-event exclusion (a detection-time choice; matches Config.freq_exclude) ---
+    # A near-ubiquitous event type links to almost everything, so it acts as a universal
+    # connector and inflates the pattern hierarchy. The default excludes any type whose mean
+    # occurrences per observation exceed 1.50 from *building* patterns (they stay in Level 0).
+    # This is consequential, so the advisor surfaces it rather than leaving it a silent default.
+    threshold = 1.50
+    totals: dict[str, int] = {}
+    for o in observations:
+        for _, ev in o.events:
+            totals[ev] = totals.get(ev, 0) + 1
+    means = {ev: totals[ev] / n_obs for ev in totals} if n_obs else {}
+    frequent = sorted((ev for ev, m in means.items() if m > threshold),
+                      key=lambda e: -means[e])
+    if frequent:
+        lst = ", ".join(f"{e} ({means[e]:.2f}/obs)" for e in frequent)
+        freq_choice = Choice(
+            "Frequent-event exclusion",
+            f"exclude {len(frequent)} type(s) from pattern-building (default)",
+            rationale=(f"{len(frequent)} event type(s) occur more than {threshold} times per "
+                       f"observation on average and are excluded by default from building "
+                       f"multi-event patterns (they remain in the Level-0 univariate set): "
+                       f"{lst}. Excluding a near-ubiquitous event stops it acting as a universal "
+                       f"connector and inflating the hierarchy; retaining it "
+                       f"(Config(exclude_events=[]), or a higher freq_exclude) keeps its "
+                       f"connector patterns at the cost of combinatorial growth. Because this "
+                       f"materially changes the pattern set, it is reported here rather than "
+                       f"applied silently."),
+            plain=("<b>Very common events are set aside from pattern-building.</b> An event that "
+                   "happens many times in almost every sequence (here: " + ", ".join(frequent) +
+                   ") links to nearly everything, so leaving it in would flood the results with "
+                   "patterns that are really just 'this common event, then anything'. By default "
+                   "we keep such events in the counts but not in the built patterns. <i>Why it "
+                   "matters:</i> it keeps the patterns meaningful; if that event is central to "
+                   "your question you can choose to keep it in."),
+            evidence={"threshold": threshold, "frequent": {e: means[e] for e in frequent}},
+        )
+    else:
+        freq_choice = Choice(
+            "Frequent-event exclusion", "none needed",
+            rationale=(f"No event type exceeds {threshold} occurrences per observation on "
+                       f"average, so none is excluded from pattern-building."),
+            plain=("No single event is common enough to swamp the patterns, so every event type "
+                   "is kept in the analysis."),
+            evidence={"threshold": threshold, "frequent": {}},
+        )
+
     return Recommendation(n_obs=n_obs, n_events=n_events,
-                          choices=[null_choice, lag_choice, err_choice])
+                          choices=[freq_choice, null_choice, lag_choice, err_choice])

{tpattern-0.1.0 → tpattern-0.1.1/tpattern.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: tpattern
-Version: 0.1.0
+Version: 0.1.1
 Summary: Open, auditable temporal & sequential pattern analysis (T-patterns) for sport and behaviour
 Author-email: Andrew Callaway <acallaway@bournemouth.ac.uk>
 License: MIT
@@ -75,7 +75,12 @@ model and clear reporting:
 
 ## Installation
 
-From source (until published on PyPI):
+```bash
+pip install tpattern
+```
+
+Add the guided-wizard extras with `pip install tpattern[gui]`. To work from source
+instead (for development or to run the reproduction scripts and shipped data):
 
 ```bash
 git clone https://github.com/ajcallaway/TPattern.git