Fanchouille / knnmodel

Create easily a fast kNN index

KNNModel

Purpose

Provide a lib to create easily a fast kNN index FastKnn use mainly nmslib as (fast) kNN backend

Install

pip install git+https://github.com/Fanchouille/knnmodel.git

Use

FastKnn builds a kNN index with specified method (default: hnsw) and space (default: cosinesimil)

• See here for different spaces
• See here for different methods

Default with hnsw method and cosinesimil space

Example with dense data:

X_train = np.random.rand(1000, 128)
y_train = np.random.randint(10, size=1000)

Unsupervised (target is id) :

from knnmodel import KNNModel
knn_model = KNNModel() 
knn_model.fit(X=X_train) # no labels
results_raw = knn_model.predict(X_test, n_neighbours=10, mode="raw", target="id") # predict & fetch list of 10 neighbours for each row in X_test
results_majority = knn_model.predict(X_test, n_neighbours=10, mode="majority_voting", target="id") # predict & fetch id with most votes pct for each row in X_test
results_on_training_set_raw = knn_model.predict_on_training_set(n_neighbours=10, mode="raw", target="id", exclude_identity=True) # predict & fetch list of 10 neighbours for each row in training set without identity
results_on_training_set_majority = knn_model.predict_on_training_set(n_neighbours=10, mode="majority_voting", target="id", exclude_identity=True)

Supervised (target is label):

from knnmodel import KNNModel    
knn_model = KNNModel()
knn_model.fit(X_train, y=y_train) # labels
results_raw = knn_model.predict(X_train, n_neighbours=10, mode="raw", target="label")
results_majority = knn_model.predict(X_train, n_neighbours=10, mode="majority_voting", target="label")
results_on_training_set_raw = knn_model.predict_on_training_set(n_neighbours=10, mode="raw", target="label")
results_on_training_set_majority = knn_model.predict_on_training_set(n_neighbours=10, mode="majority_voting", target="label")

Development

Clone project

Install Anaconda local environment as below:

./install.sh

Activate Anaconda local environment as below:

conda activate ${PWD}/.conda