knn_missing_data.py 文件源码

def knn_masked_data(trX,trY,missing_data_dir, input_shape, k):

    raw_im_data = np.loadtxt(join(script_dir,missing_data_dir,'index.txt'),delimiter=' ',dtype=str)
    raw_mask_data = np.loadtxt(join(script_dir,missing_data_dir,'index_mask.txt'),delimiter=' ',dtype=str)
    # Using 'brute' method since we only want to do one query per classifier
    # so this will be quicker as it avoids overhead of creating a search tree
    knn_m = KNeighborsClassifier(algorithm='brute',n_neighbors=k)
    prob_Y_hat = np.zeros((raw_im_data.shape[0],int(np.max(trY)+1)))
    total_images = raw_im_data.shape[0]
    pbar = progressbar.ProgressBar(widgets=[progressbar.FormatLabel('\rProcessed %(value)d of %(max)d Images '), progressbar.Bar()], maxval=total_images, term_width=50).start()
    for i in range(total_images):
        mask_im=load_image(join(script_dir,missing_data_dir,raw_mask_data[i][0]), input_shape,1).reshape(np.prod(input_shape))
        mask = np.logical_not(mask_im > eps) # since mask is 1 at missing locations
        v_im=load_image(join(script_dir,missing_data_dir,raw_im_data[i][0]), input_shape, 255).reshape(np.prod(input_shape))
        rep_mask = np.tile(mask,(trX.shape[0],1))
        # Corrupt whole training set according to the current mask
        corr_trX = np.multiply(trX, rep_mask)        
        knn_m.fit(corr_trX, trY)
        prob_Y_hat[i,:] = knn_m.predict_proba(v_im.reshape(1,-1))
        pbar.update(i)
    pbar.finish()
    return prob_Y_hat