reranker first version

README.md

@@ -1,26 +1,27 @@
-# Cranker
-
-[![awesome plugin](https://custom-icon-badges.demolab.com/static/v1?label=&message=awesome+plugin&color=383938&style=for-the-badge&logo=cheshire_cat_ai)](https://)  
-[![Awesome plugin](https://custom-icon-badges.demolab.com/static/v1?label=&message=Awesome+plugin&color=000000&style=for-the-badge&logo=cheshire_cat_ai)](https://)  
+# Cheshire Cat ReRanker
+
 [![awesome plugin](https://custom-icon-badges.demolab.com/static/v1?label=&message=awesome+plugin&color=F4F4F5&style=for-the-badge&logo=cheshire_cat_black)](https://)
 
-Write here all the useful information about your plugin.
+Cheshire Cat ReRanker apply a rearrangement at each memory with different criteria:
+* the [*episodic memory*](https://cheshire-cat-ai.github.io/docs/conceptual/memory/episodic_memory/) is reordered according to a temporal criteria, from the newest to the oldest.
+* the [*declarative memory*](https://cheshire-cat-ai.github.io/docs/conceptual/memory/declarative_memory/) is reordered according the [**lost in the middle** paper](https://arxiv.org/abs/2307.03172) method
+* the [*procedural memory*](https://cheshire-cat-ai.github.io/docs/conceptual/memory/procedural_memory/) isn't reordered but filtered using a threshold, the default value is 0.5 but you can set it.
 
-This repository is the template to automate the release of official Cheshire Cat AI plugins. 
+The ReRanker can be enabled and disabled for each memory independently using the "Settings".
 
-## Usage
+## Installation
 
-1. Create a new repository clicking on the `Use this template` button.
-2. Clone your new repo directly in the Cat's `plugins` folder.
-3. Run the `setup.py` script:
-```bash
-python setup.py
-```
-The script will prompt you to write the name of your plugin and make an initial setup setting the name in the files.
+1. Navigate to the `Plugins` page;
+2. Scroll until you find the "Cheshire Cat ReRanker" plugin;
+3. Click on `Install
 
-4. Start developing!
+## Usage
 
-> **Important**
-> A new release of your plugin is triggered every time you set a new `version` in the `plugin.json` file.
-> Please, remember to set it correctly every time to want to release an update.
+1. Navigate to the `Plugins` page;
+2. Scroll until you find the "Cheshire Cat ReRanker" plugin;
+3. Click on the "Settings" icon on the bottom right of the plugin card;
+4. Edit the available settings and save.
 
+> **Important**
+> ReRankers (in particular Lost In The Middle) are very useful if you get at least more than 10 documents returned from the memories.
+> Before using it download and enable the [C.A.T. plugin](https://github.com/Furrmidable-Crew/cat_advanced_tools) from the Plugins store, [follow the instructions](https://github.com/Furrmidable-Crew/cat_advanced_tools#usage) to increase the k parameter of the memories.

cranker.py → ccat_reranker.py

@@ -1,5 +1,5 @@
 from cat.mad_hatter.decorators import hook
-from .rankers import litm, get_settings, recentness
+from .rankers import get_settings, recent_ranker, litm, filter_ranker
 
 
 @hook(priority=1)
@@ -18,7 +18,7 @@ def after_cat_recalls_memories(cat) -> None:
     settings = get_settings()
     if settings["RECENTNESS"]:
         if cat.working_memory['episodic_memories']:
-            recent_docs = recentness(cat.working_memory['episodic_memories'])
+            recent_docs = recent_ranker(cat.working_memory['episodic_memories'])
             cat.working_memory['episodic_memories'] = recent_docs
         else:
             print("#HicSuntGattones")
@@ -32,7 +32,7 @@ def after_cat_recalls_memories(cat) -> None:
 
     if settings["FILTER"]:
         if cat.working_memory['procedural_memories']:
-            print("NON PUò ENTRAREEEEH!")
+            filtered = filter_ranker(cat.working_memory['procedural_memories'], settings["tool_threshold"])
         else:
             print("#HicSuntGattones")
     pass # do nothing

plugin.json

@@ -1,10 +1,9 @@
 {
     "name": "Cheshire Cat Reranker",
-    "version": "0.0.1",
-    "description": "Description of cranker.",
+    "version": "0.1.0",
+    "description": "This plugin apply a reranking to each memory. It rearranges: the episodic memory from the most recent to the oldest, the declarative using lost in the middle and the procedural using a filter.",
     "author_name": "nickprock",
-    "author_url": "https://mywebsite.me",
-    "plugin_url": "https://github.com/my_name/cranker",
-    "tags": "cat, template, example",
-    "thumb": "https://raw.githubusercontent.com/my_repo_path/cranker.png"
+    "plugin_url": "https://github.com/nickprock/ccat_reranker",
+    "tags": "cat, memory, advanced, reranker",
+    "thumb": "https://raw.githubusercontent.com/nickprock/ccat_reranker/main/img.jpg"
 }

rankers.py

@@ -53,7 +53,7 @@ def litm(documents):
         litm_docs = [documents[idx] for idx in lost_in_the_middle_indices]
     return litm_docs
 
-def filter_ranker(documents):
+def filter_ranker(documents, tool_threshold):
     """
     Returns list of tools with a similarity score higher than 0.5.
 
@@ -63,5 +63,7 @@ def filter_ranker(documents):
 
     Returns
     ----------
-    tools: The same list but reordered
-    """
+    filtered: The same list but filtered
+    """
+    filtered = [d for d in documents if d[1]>tool_threshold]
+    return filtered

settings.py

@@ -0,0 +1,13 @@
+from cat.mad_hatter.decorators import plugin
+from pydantic import BaseModel, Field
+
+
+class MySettings(BaseModel):
+    LITM: bool = True,
+    RECENTNESS: bool = True,
+    FILTER: bool = True,
+    tool_threshold: float = 0.5
+
+@plugin
+def settings_schema():
+    return MySettings.schema()