gistfile1.txt

from common import *
from ventilator import *
import faiss

data_dir = root_dir+'/data'
train_df = pd.read_csv(data_dir + '/train.csv')
test_df  = pd.read_csv(data_dir + '/test.csv')


if 1:  #check statistics
    for r in [5,20,50]:
        for c in [10,20,50]:
            t_df = train_df[(train_df.R == r) & (train_df.C == c)]
            t_num   = len(t_df)//80
            t_u_out = t_df.u_out.values.reshape(-1, 80)
            t_u_in  = t_df.u_in.values.reshape(-1, 80)
            t_p     = t_df.pressure.values.reshape(-1,80)
            t_t     = t_df.time_step.values.reshape(-1,80)


            e_df = test_df[(test_df.R == r) & (test_df.C == c)]
            e_num   = len(e_df)//80
            e_u_out = e_df.u_out.values.reshape(-1, 80)
            e_u_in  = e_df.u_in.values.reshape(-1, 80)
            #e_p     = e_df.pressure.values.reshape(-1,80)
            e_t     = e_df.time_step.values.reshape(-1,80)

            #---
            dim = 80  # dimension
            nb = t_num  # database size
            nq = e_num  # nb of queries
            xb = t_u_in.astype('float32')
            xq = e_u_in.astype('float32')

            #https://github.com/facebookresearch/faiss/blob/main/tutorial/python/4-GPU.py
            res = faiss.StandardGpuResources()  # use a single GPU
            index_flat = faiss.IndexFlatL2(dim)
            gpu_index_flat = faiss.index_cpu_to_gpu(res, 0, index_flat)

            gpu_index_flat.add(xb)  # add vectors to the index
            print(gpu_index_flat.ntotal)

            num_neighbour = 5  # we want to see 5 nearest neighbors
            D, I = gpu_index_flat.search(xq, num_neighbour)  # actual search

            for i in range(e_num):
                plt.clf()
                plt.plot(e_t[i], e_u_in[i], lw=2, alpha=1, label='test u_in',c='black')

                plt.plot(t_t[I[i,0]], t_u_in[I[i,0]], '--', lw=2, alpha=1, label='knn-1 u_in %4.1f'%D[i,0],c='green')
                plt.plot(t_t[I[i,1]], t_u_in[I[i,1]], '--', lw=1, alpha=1, label='knn-2 u_in %4.1f'%D[i,1],c='green')
                plt.plot(t_t[I[i,2]], t_u_in[I[i,2]], '--', lw=1, alpha=0.25, label='knn-3 u_in %4.1f'%D[i,2],c='green')


                plt.plot(t_t[I[i,0]], t_p[I[i,0]], '--', lw=2, alpha=1, label='knn-1 press %4.1f'%D[i,0],c='red')
                plt.plot(t_t[I[i,1]], t_p[I[i,1]], '--', lw=1, alpha=1, label='knn-2 press %4.1f'%D[i,1],c='red')
                plt.plot(t_t[I[i,2]], t_p[I[i,2]], '--', lw=1, alpha=0.25, label='knn-3 press %4.1f'%D[i,2],c='red')

                plt.ylim([-5, 65])
                plt.xlim([0, 2.5])
                plt.legend()
                # plt.show()
                while not plt.waitforbuttonpress(): pass

            zz=0