app.py

import gradio as gr
import torch
import numpy as np
import h5py
import faiss
from PIL import Image
import io 
import pickle
import random

def getRandID():
    indx = random.randrange(0, 396503)
    return indx_to_id_dict[indx], indx

def chooseImageIndex(indexType):
    if (indexType == "FlatIP(default)"):
        return image_index_IP
    elif (indexType == "FlatL2"):
        return image_index_L2
    elif (indexType == "HNSWFlat"):
        return image_index_HNSW
    elif (indexType == "IVFFlat"):
        return image_index_IVF
    elif (indexType == "LSH"):
        return image_index_LSH

def chooseDNAIndex(indexType):
    if (indexType == "FlatIP(default)"):
        return dna_index_IP
    elif (indexType == "FlatL2"):
        return dna_index_L2
    elif (indexType == "HNSWFlat"):
        return dna_index_HNSW
    elif (indexType == "IVFFlat"):
        return dna_index_IVF
    elif (indexType == "LSH"):
        return dna_index_LSH



def searchEmbeddings(id, mod1, mod2, indexType):
    # variable and index initialization
    dim = 768
    count = 0
    num_neighbors = 10

    index = faiss.IndexFlatIP(dim)

    # get index
    if (mod2 == "Image"):
        index = chooseImageIndex(indexType)
    elif (mod2 == "DNA"):
        index = chooseDNAIndex(indexType)
    

    # search for query
    if (mod1 == "Image"):
        query = id_to_image_emb_dict[id]
    elif (mod1 == "DNA"):
        query = id_to_dna_emb_dict[id]
    query = query.astype(np.float32)
    D, I = index.search(query, num_neighbors)

    id_list = []
    i = 1
    for indx in I[0]:
        id = indx_to_id_dict[indx]
        id_list.append(id)
        
    return id_list

with gr.Blocks() as demo:

    # for hf: change all file paths, indx_to_id_dict as well

    # load indexes
    image_index_IP = faiss.read_index("big_image_index_FlatIP.index")
    image_index_L2 = faiss.read_index("big_image_index_FlatL2.index")
    image_index_HNSW = faiss.read_index("big_image_index_HNSWFlat.index")
    image_index_IVF = faiss.read_index("big_image_index_IVFFlat.index")
    image_index_LSH = faiss.read_index("big_image_index_LSH.index")

    dna_index_IP = faiss.read_index("big_dna_index_FlatIP.index")
    dna_index_L2 = faiss.read_index("big_dna_index_FlatL2.index")
    dna_index_HNSW = faiss.read_index("big_dna_index_HNSWFlat.index")
    dna_index_IVF = faiss.read_index("big_dna_index_IVFFlat.index")
    dna_index_LSH = faiss.read_index("big_dna_index_LSH.index")

    with open("dataset_processid_list.pickle", "rb") as f:
        dataset_processid_list = pickle.load(f)
    with open("processid_to_index.pickle", "rb") as f:
        processid_to_index = pickle.load(f)
    with open("big_indx_to_id_dict.pickle", "rb") as f:
        indx_to_id_dict = pickle.load(f)

    # initialize both possible dicts
    with open("big_id_to_image_emb_dict.pickle", "rb") as f:
        id_to_image_emb_dict = pickle.load(f)
    with open("big_id_to_dna_emb_dict.pickle", "rb") as f:
        id_to_dna_emb_dict = pickle.load(f)

    with gr.Column():
        with gr.Row():
            with gr.Column():
                rand_id = gr.Textbox(label="Random ID:")
                rand_id_indx = gr.Textbox(label="Index:")
                id_btn = gr.Button("Get Random ID")
            with gr.Column():
                mod1 = gr.Radio(choices=["DNA", "Image"], label="Search From:")
                mod2 = gr.Radio(choices=["DNA", "Image"], label="Search To:")

        indexType = gr.Radio(choices=["FlatIP(default)", "FlatL2", "HNSWFlat", "IVFFlat", "LSH"], label="Index:", value="FlatIP(default)")
        process_id = gr.Textbox(label="ID:", info="Enter a sample ID to search for")
        process_id_list = gr.Textbox(label="Closest 10 matches:" )
        search_btn = gr.Button("Search") 
        id_btn.click(fn=getRandID, inputs=[], outputs=[rand_id, rand_id_indx])

    search_btn.click(fn=searchEmbeddings, inputs=[process_id, mod1, mod2, indexType], 
                     outputs=[process_id_list])
    

demo.launch()