0

I am using ScaNN to perform similarity searches and would like to place more emphasis on some features than others when performing a similarity search.

for example, if I have the following data

name  | age | country | income
John     29     US        $47k
Susan    28     US        $44k
Bill     26     US        $39k
Sarah    35     UK        $100k
Jack     34     UK        $90k
Maggie   37     UK        $95k

and income has more importance, then given the following query:

George, 28, US, $100k

it would return

Sarah, Jack, Maggie

adding more weight to the income feature.

Training data values are normalized before building the similarity index

df_np = preprocessing.normalize(df[features])

and likewise the query values are normalized before performing a search

np_q = preprocessing.normalize([list(query.values())])
user4157124
  • 2,809
  • 13
  • 27
  • 42
noobie
  • 2,427
  • 5
  • 41
  • 66

0 Answers0