chore: create faiss package

tasks/qualification/CMakeLists.txt → common/vector_databases/lasr_vector_databases_faiss/CMakeLists.txt

@@ -1,5 +1,5 @@
 cmake_minimum_required(VERSION 3.0.2)
-project(qualification)
+project(lasr_vector_databases_faiss)
 
 ## Compile as C++11, supported in ROS Kinetic and newer
 # add_compile_options(-std=c++11)

tasks/qualification/package.xml → common/vector_databases/lasr_vector_databases_faiss/package.xml

@@ -1,13 +1,13 @@
 <?xml version="1.0"?>
 <package format="2">
-  <name>qualification</name>
+  <name>lasr_vector_databases_faiss</name>
   <version>0.0.0</version>
-  <description>The qualification package</description>
+  <description>The faiss package</description>
 
   <!-- One maintainer tag required, multiple allowed, one person per tag -->
   <!-- Example:  -->
   <!-- <maintainer email="[email protected]">Jane Doe</maintainer> -->
-  <maintainer email="jared.swift@kcl.ac.uk">Jared Swift</maintainer>
+  <maintainer email="matthew.s.barker@kcl.ac.uk">Matt Barker</maintainer>
 
 
   <!-- One license tag required, multiple allowed, one license per tag -->
@@ -50,8 +50,6 @@
   <!--   <doc_depend>doxygen</doc_depend> -->
   <buildtool_depend>catkin</buildtool_depend>
   <build_depend>catkin_virtualenv</build_depend>
-  <depend>lasr_vision_msgs</depend>
-  <depend>lasr_speech_recognition_msgs</depend>
   <build_depend>message_generation</build_depend>
   <exec_depend>message_runtime</exec_depend>
   <build_depend>actionlib_msgs</build_depend>

common/vector_databases/lasr_vector_databases_faiss/requirements.in

@@ -0,0 +1 @@
+faiss-cpu

tasks/qualification/requirements.txt → common/vector_databases/lasr_vector_databases_faiss/requirements.txt

@@ -4,7 +4,6 @@
 certifi==2024.2.2         # via requests
 charset-normalizer==3.3.2  # via requests
 click==8.1.7              # via nltk
-clip @ git+https://github.com/openai/CLIP.git  # via -r requirements.in
 faiss-cpu==1.7.4          # via -r requirements.in
 filelock==3.13.1          # via huggingface-hub, torch, transformers, triton
 fsspec==2024.2.0          # via huggingface-hub, torch

tasks/qualification/setup.py → common/vector_databases/lasr_vector_databases_faiss/setup.py

@@ -4,7 +4,7 @@
 from catkin_pkg.python_setup import generate_distutils_setup
 
 setup_args = generate_distutils_setup(
-    packages=["qualification"], package_dir={"": "src"}
+    packages=["lasr_vector_databases_faiss"], package_dir={"": "src"}
 )
 
 setup(**setup_args)

tasks/qualification/src/qualification/command_similarity.py → common/vector_databases/lasr_vector_databases_faiss/src/qualification/command_similarity.py

@@ -3,9 +3,6 @@
 import torch
 import numpy as np
 
-# import matplotlib.pyplot as plt
-from sklearn.manifold import TSNE
-
 # import rospy
 import faiss  # type: ignore
 from sentence_transformers import SentenceTransformer  # type: ignore
@@ -57,11 +54,18 @@ def get_sentence_embeddings(
     )
 
 
-def create_vector_database(command_embeddings: np.ndarray) -> faiss.IndexFlatIP:
+def create_vector_database(vectors: np.ndarray) -> faiss.IndexFlatIP:
+    """Creates a vector database from an array of vectors of the same dimensionality
+    Args:
+        vectors (np.ndarray): shape (n_vectors, vector_dim)
+
+    Returns:
+        faiss.IndexFlatIP: Flat index containing the vectors
+    """
     print("Creating vector database")
-    index_flat = faiss.IndexFlatIP(command_embeddings.shape[1])
-    faiss.normalize_L2(command_embeddings)
-    index_flat.add(command_embeddings)
+    index_flat = faiss.IndexFlatIP(vectors.shape[1])
+    faiss.normalize_L2(vectors)
+    index_flat.add(vectors)
     print("Finished creating vector database")
     return index_flat
 
@@ -141,6 +145,7 @@ def get_similar_commands(
 
 
 if __name__ == "__main__":
+    """Example usage of using this to find similar commands"""
     command = "find Jared and asks if he needs help"
     result, distances, command_embeddings, query_embedding = get_similar_commands(
         command,
@@ -149,47 +154,4 @@ def get_similar_commands(
         n_similar_commands=1000,
         return_embeddings=True,
     )
-    command_embeddings = np.concatenate((command_embeddings, query_embedding))
-    # tsne = TSNE(
-    #     n_components=2,
-    #     perplexity=3,
-    #     verbose=1,
-    #     n_iter=1000,
-    #     n_iter_without_progress=1000,
-    # )
-    # print("Fitting tsne")
-    # tsne_embed = tsne.fit_transform(command_embeddings)
-    # # plot the embeddings
-    # plt.scatter(tsne_embed[:-1, 0], tsne_embed[:-1, 1], label="Commands")
-    # plt.scatter(tsne_embed[-1, 0], tsne_embed[-1, 1], label="Query")
-    # # Annotate the query
-    # plt.annotate(
-    #     command,
-    #     (tsne_embed[-1, 0], tsne_embed[-1, 1]),
-    # )
-    # # Set title
-    # plt.title("t-SNE plot of 1000 closest command embeddings")
-    # # Label axes
-    # plt.xlabel("t-SNE Dimension 1")
-    # plt.ylabel("t-SNE Dimension 2")
-    # print(f"Command: {command}")
-    # for i in range(3):
-    #     # decide if it should be th or st
-    #     if i == 0:
-    #         suffix = "st"
-    #     elif i == 1:
-    #         suffix = "nd"
-    #     elif i == 2:
-    #         suffix = "rd"
-    #     else:
-    #         suffix = "th"
-    #     print(
-    #         f"The {i+1}{suffix} closest command is {result[i]} with a similarity of {distances[i]:.2f}"
-    #     )
-    #     # Annotate the points
-    #     plt.annotate(
-    #         result[i],
-    #         (tsne_embed[i, 0], tsne_embed[i, 1]),
-    #     )
-
-    # plt.show()
+    print(result)

common/vision/lasr_vision_clip/setup.py

@@ -4,7 +4,7 @@
 from catkin_pkg.python_setup import generate_distutils_setup
 
 setup_args = generate_distutils_setup(
-    packages=["qualification"], package_dir={"": "src"}
+    packages=["lasr_vision_clip"], package_dir={"": "src"}
 )
 
 setup(**setup_args)