Clustering

[jerinphilip]

x = img | prediction | (img, prediction)                                
i = index of occurunce                                                  

X = [(x, i)]                                                            

d = lambda (x_1, i_1), (x_2, i_2): some-distance(x_1, x2)               

# cluster : predictions, distance-metric -> [[predictions]]             
def cluster(X, d, method):                                              

def MSTCluster(X, d):                                                    
# https://github.com/jakevdp/mst_clustering/blob/master/mst_clustering/_mst_clustering.py

def LSHCluster(X, d):                                                    

[jerinphilip]

Start with praveen sir's code and pipe it to the cluster function, already written by now.

Use the function here to get features and connect it to cluster with a euclidean distance criteria. Manually try parameters threshold and prune for now and see if anything works okay.

The features are in /OCRData2/praveen-intermediate/<book-id>/feats.npy, for the respective book. To test the function, you can simply run

python3 -m doctools.hwnet.codes.deploy.feature -f /OCRData2/praveen-intermediate/0061/feats.npy

@Deepayan137

The glue code we'll reuse, hyperparameter tuning, we'll do later.