Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Changes in Layouts , Addition of new files #1000

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Changes in Layouts , Addition of new files #1000

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 36 additions & 7 deletions umap/layouts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
import numpy as np
import numba
import umap.distances as dist
import umap_extend
from umap.utils import tau_rand_int
from tqdm.auto import tqdm

Expand Down Expand Up @@ -346,7 +347,6 @@ def optimize_layout_euclidean(
tqdm_kwds["disable"] = not verbose

for n in tqdm(range(n_epochs), **tqdm_kwds):

densmap_flag = (
densmap
and (densmap_kwds["lambda"] > 0)
Expand Down Expand Up @@ -376,7 +376,7 @@ def optimize_layout_euclidean(
dens_re_mean = 0
dens_re_cov = 0

optimize_fn(
umap_extend.cpp_optimize_layout_euclidean_single_epoch(
head_embedding,
tail_embedding,
head,
Expand Down Expand Up @@ -405,13 +405,42 @@ def optimize_layout_euclidean(
dens_mu,
dens_mu_tot,
)

alpha = initial_alpha * (1.0 - (float(n) / float(n_epochs)))

if verbose and n % int(n_epochs / 10) == 0:
# optimize_fn(
# head_embedding,
# tail_embedding,
# head,
# tail,
# n_vertices,
# epochs_per_sample,
# a,
# b,
# rng_state,
# gamma,
# dim,
# move_other,
# alpha,
# epochs_per_negative_sample,
# epoch_of_next_negative_sample,
# epoch_of_next_sample,
# n,
# densmap_flag,
# dens_phi_sum,
# dens_re_sum,
# dens_re_cov,
# dens_re_std,
# dens_re_mean,
# dens_lambda,
# dens_R,
# dens_mu,
# dens_mu_tot,
# )

alpha = initial_alpha * (1.0 - (float(n) / float(n_epochs)))

if verbose and n % int(n_epochs / 10) == 0:
print("\tcompleted ", n, " / ", n_epochs, "epochs")

if epochs_list is not None and n in epochs_list:
if epochs_list is not None and n in epochs_list:
embedding_list.append(head_embedding.copy())

# Add the last embedding to the list as well
Expand Down
241 changes: 241 additions & 0 deletions umap_extend.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,241 @@
/**********************************************************************
MIT License

Copyright (c) 2022 Intel Labs

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Authors: Narendra Chaudhary <[email protected]>; Sanchit Misra <[email protected]>
*****************************************************************************************/



#include<pybind11/pybind11.h>
#include<pybind11/numpy.h>
#include<fstream>
#include<iostream>
#include<omp.h>
#include<math.h>
#include <x86intrin.h>
#include "xsrandom.hpp"

namespace py = pybind11;

// def _optimize_layout_euclidean_single_epoch(
// head_embedding, // <class 'numpy.ndarray'>, dtype = float32, shape = (990, 2)
// tail_embedding, // <class 'numpy.ndarray'>, dtype = float32, shape = (990, 2)
// head, // <class 'numpy.ndarray'>, dtype = int, shape = (22116,)
// tail, // <class 'numpy.ndarray'>, dtype = int, shape = (22116,)
// n_vertices, // <class 'int'>
// epochs_per_sample, // <class 'numpy.ndarray'>, dtype = float64, shape = (22116,)
// a, // <class 'numpy.float64'>
// b, // <class 'numpy.float64'>
// rng_state, // <class 'numpy.ndarray'>, dtype = int64, shape = (3,)
// gamma, // <class 'float'>
// dim, // <class 'int'>
// move_other, // <class 'bool'>
// alpha, // <class 'float'>
// epochs_per_negative_sample, // <class 'numpy.ndarray'>, dtype = float64, shape = (22116,)
// epoch_of_next_negative_sample, // <class 'numpy.ndarray'>, dtype = float64, shape = (22116,)
// epoch_of_next_sample, // <class 'numpy.ndarray'>, dtype = float64, shape = (22116,)
// n, // <class 'int'>
// densmap_flag, // <class 'bool'>
// dens_phi_sum, // <class 'numpy.ndarray'>, dtype = float32, shape = (1,)
// dens_re_sum, // <class 'numpy.ndarray'>, dtype = float32, shape = (1,)
// dens_re_cov, // <class 'int'>
// dens_re_std, // <class 'int'>
// dens_re_mean, // <class 'int'>
// dens_lambda, // <class 'int'>
// dens_R, // <class 'numpy.ndarray'>, dtype = float32, shape = (1,)
// dens_mu, // <class 'numpy.ndarray'>, dtype = float32, shape = (1,)
// dens_mu_tot, // <class 'int'>
// )

inline double clip(double val){
if(val > 4.0)
return 4.0;
else if(val < -4.0)
return -4.0;
else
return val;
}

// long tau_rand_int(long *state){
// state[0] = (((state[0] & 4294967294) << 12) & 0xFFFFFFFF) ^ ((((state[0] << 13) & 0xFFFFFFFF) ^ state[0]) >> 19);
// state[1] = (((state[1] & 4294967288) << 4) & 0xFFFFFFFF) ^ ((((state[1] << 2) & 0xFFFFFFFF) ^ state[1]) >> 25);
// state[2] = (((state[2] & 4294967280) << 17) & 0xFFFFFFFF) ^ ((((state[2] << 3) & 0xFFFFFFFF) ^ state[2]) >> 11);

// return int(state[0] ^ state[1] ^ state[2]);
// }

void cpp_optimize_layout_euclidean_single_epoch(py::array_t<float> head_embedding, py::array_t<float> tail_embedding, py::array_t<int> head, py::array_t<int> tail, \
int n_vertices, py::array_t<double> epochs_per_sample, double a, double b, py::array_t<long> rng_state, float gamma, int dim, bool move_other, \
float alpha, py::array_t<double> epochs_per_negative_sample, py::array_t<double> epoch_of_next_negative_sample, py::array_t<double> epoch_of_next_sample, \
int n, bool densmap_flag, py::array_t<float> dens_phi_sum, py::array_t<float> dens_re_sum, int dens_re_cov, int dens_re_std, int dens_re_mean, int dens_lambda, \
py::array_t<float> dens_R, py::array_t<float> dens_mu, int dens_mu_tot){

py::buffer_info buf_head_embedding = head_embedding.request();
py::buffer_info buf_tail_embedding = tail_embedding.request();
py::buffer_info buf_head = head.request();
py::buffer_info buf_tail = tail.request();

py::buffer_info buf_epochs_per_sample = epochs_per_sample.request();
// py::buffer_info buf_rng_state = rng_state.request();

py::buffer_info buf_epochs_per_negative_sample = epochs_per_negative_sample.request();
py::buffer_info buf_epoch_of_next_negative_sample = epoch_of_next_negative_sample.request();
py::buffer_info buf_epoch_of_next_sample = epoch_of_next_sample.request();

py::buffer_info buf_dens_phi_sum = dens_phi_sum.request();
py::buffer_info buf_dens_re_sum = dens_re_sum.request();
py::buffer_info buf_dens_R = dens_R.request();
py::buffer_info buf_dens_mu = dens_mu.request();
cout<<"Entered the CPP file !";
float *head_embedding_ptr = (float *)buf_head_embedding.ptr;
float *tail_embedding_ptr = (float *)buf_tail_embedding.ptr;
int *head_ptr = (int *)buf_head.ptr;
int *tail_ptr = (int *)buf_tail.ptr;

double *epochs_per_sample_ptr = (double *)buf_epochs_per_sample.ptr;
// long *rng_state_ptr = (long *)buf_rng_state.ptr;

double *epochs_per_negative_sample_ptr = (double *)buf_epochs_per_negative_sample.ptr;
double *epoch_of_next_negative_sample_ptr = (double *)buf_epoch_of_next_negative_sample.ptr;
double *epoch_of_next_sample_ptr = (double *)buf_epoch_of_next_sample.ptr;

float *dens_phi_sum_ptr = (float *)buf_dens_phi_sum.ptr;
float *dens_re_sum_ptr = (float *)buf_dens_re_sum.ptr;
float *dens_R_ptr = (float *)buf_dens_R.ptr;
float *dens_mu_ptr = (float *)buf_dens_mu.ptr;

// #pragma omp parallel for schedule(dynamic, 1000)
#pragma omp parallel
{
#pragma omp for
for(int i=0; i < buf_epochs_per_sample.shape[0]; i++){ // 23029392
if(epoch_of_next_sample_ptr[i] <= n){ // 6928928
int j = head_ptr[i]; // (Size = 4*6928928)
int k = tail_ptr[i]; // (Size = 4*6928928)

float *current, *other;
current = &head_embedding_ptr[j*dim]; // (Size = 16*990000)
other = &tail_embedding_ptr[k*dim]; // (Size = 16*990000)

// for(int d = 0; d < dim; d++){
// current[d] = head_embedding_ptr[j*dim + d];
// other[d] = tail_embedding_ptr[k*dim + d];
// }

float dist_squared=0.0f;
for(int d = 0; d < dim; d++)
dist_squared += ((current[d] - other[d]) * (current[d] - other[d]));

double grad_cor_coeff=0.0;
if(densmap_flag){
double phi = 1.0 / (1.0 + a * pow(dist_squared, b));
double dphi_term = (a * b * pow(dist_squared, b - 1) / (1.0 + a * pow(dist_squared, b)));

double q_jk = phi / dens_phi_sum_ptr[k];
double q_kj = phi / dens_phi_sum_ptr[j];

double drk = q_jk * ((1.0 - b * (1 - phi)) / exp(dens_re_sum_ptr[k]) + dphi_term);
double drj = q_kj * ((1.0 - b * (1 - phi)) / exp(dens_re_sum_ptr[j]) + dphi_term);

double re_std_sq = dens_re_std * dens_re_std;

double weight_k = (dens_R_ptr[k] - dens_re_cov * (dens_re_sum_ptr[k] - dens_re_mean) / re_std_sq);
double weight_j = (dens_R_ptr[j] - dens_re_cov * (dens_re_sum_ptr[j] - dens_re_mean) / re_std_sq);

grad_cor_coeff = (
dens_lambda
* dens_mu_tot
* (weight_k * drk + weight_j * drj)
/ (dens_mu_ptr[i] * dens_re_std)
/ n_vertices
);
}

double grad_coeff=0.0;
if(dist_squared > 0.0){
grad_coeff = -2.0 * a * b * pow(dist_squared, b - 1.0);
grad_coeff /= (a * pow(dist_squared, b) + 1.0);
}
else
grad_coeff = 0.0;

double grad_d;
for(int d = 0; d < dim; d++){
grad_d = clip(grad_coeff * (current[d] - other[d]));

if(densmap_flag)
grad_d += clip(2.0 * grad_cor_coeff * (current[d] - other[d]));

current[d] += (grad_d * alpha);
if(move_other)
other[d] += (-grad_d * alpha);
}

epoch_of_next_sample_ptr[i] += epochs_per_sample_ptr[i]; // (Size = 2*8*6928928)

int n_neg_samples = (int)((n - epoch_of_next_negative_sample_ptr[i]) / epochs_per_negative_sample_ptr[i]); // (Size = 2*8*6928928)

XSRandom prng;
unsigned int list[(n_neg_samples/16 + 1)*16];
prng.fillArray(list, (n_neg_samples/16 + 1)*16);
for(int p = 0; p < n_neg_samples; p++){ // 5 times on average

// k = tau_rand_int(rng_state_ptr) % n_vertices; // ((n % M) + M) % M
// k = ((tau_rand_int(rng_state_ptr) % n_vertices) + n_vertices) % n_vertices; // ((n % M) + M) % M
k = list[p] % n_vertices;

other = &tail_embedding_ptr[k*dim];

dist_squared=0.0f;
for(int d = 0; d < dim; d++)
dist_squared += ((current[d] - other[d]) * (current[d] - other[d]));

if(dist_squared > 0.0){
grad_coeff = 2.0 * gamma * b;
grad_coeff /= ((0.001 + dist_squared) * (a * pow(dist_squared, b) + 1.0));
}
else if(j == k)
continue;
else
grad_coeff = 0.0;

for(int d = 0; d < dim; d++){
if(grad_coeff > 0.0)
grad_d = clip(grad_coeff * (current[d] - other[d]));
else
grad_d = 4.0;
current[d] += (grad_d * alpha);
}
}

epoch_of_next_negative_sample_ptr[i] += (n_neg_samples * epochs_per_negative_sample_ptr[i]);
}
}
}
}


PYBIND11_MODULE(umap_extend, handle){
handle.doc() = "This is the module for _optimize_layout_euclidean_single_epoch() function.";
handle.def("cpp_optimize_layout_euclidean_single_epoch", &cpp_optimize_layout_euclidean_single_epoch);
}
Loading