bipartite

Learning Bipartite Graphs: Heavy Tails and Multiple Components (NeurIPS 2022)

https://github.com/convexfi/bipartite

Repository

Learning Bipartite Graphs: Heavy Tails and Multiple Components (NeurIPS 2022)

README.Rmd

---
title: "bipartite README"
output:
  html_document:
    keep_md: true
---

```{r, echo = FALSE}
library(knitr)
opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  fig.align = "center",
  fig.retina = 2,
  out.width = "75%",
  dpi = 96
)
knit_hooks$set(pngquant = hook_pngquant)
```

# finbipartite

This repo contains numerical implementations of algorithms to estimate weighted, undirected, (possibly k-component) bipartite graphs.


## Installation

**finbipartite** depends on the development version of **spectralGraphTopology**.

You can install the development version from GitHub:
```{r, eval = FALSE}
> devtools::install_github("convexfi/spectralGraphTopology")
> devtools::install_github("convexfi/finbipartite")
```

#### Microsoft Windows
On MS Windows environments, make sure to install the most recent version of ``Rtools``.

## Usage

### Clustering a network of S&P500 stocks
```{r plot_sp500_stocks_network, message=FALSE}
library(fitHeavyTail)
library(xts)
library(quantmod)
library(igraph)
library(finbipartite)
library(readr)
set.seed(42)

# load SP500 stock prices into an xts table
stock_prices <- readRDS("examples/stocks/sp500-data-2016-2021.rds")
# number of sectors
q <- 8
# number of stocks
r <- ncol(stock_prices) - q
# total nodes in the graph
p <- r + q
colnames(stock_prices)[1:r]
colnames(stock_prices)[(r+1):p]
selected_sectors <- c("Consumer Staples", "Energy", "Financials",
                      "Health Care", "Industrials", "Materials", 
                      "Real Estate", "Utilities")
# compute log-returns
log_returns <- diff(log(stock_prices), na.pad = FALSE)
# fit an univariate Student-t distribution to the log-returns of the market
# to obtain an estimate of the degrees of freedom (nu)
mkt_index <- Ad(getSymbols("^GSPC",
              from = index(stock_prices[1]), to = index(stock_prices[nrow(stock_prices)]), 
              auto.assign = FALSE, 
              verbose = FALSE))
mkt_index_log_returns <- diff(log(mkt_index), na.pad = FALSE)
nu <- fit_mvt(mkt_index_log_returns,
              nu="MLE-diag-resampled")$nu
# learn a bipartite graph with k = 8 components
graph_mrf <- learn_heavy_tail_kcomp_bipartite_graph(scale(log_returns),
                                                    r = r,
                                                    q = q,
                                                    k = 8,
                                                    nu = nu,
                                                    learning_rate = 1,
                                                    verbose = FALSE)
# save predicted labels
labels_pred <- c()
for(i in c(1:r)){
    labels_pred <- c(labels_pred, which.max(graph_mrf$B[i, ]))
}
# build network
SP500 <- read_csv("examples/stocks/SP500-sectors.csv")
stock_sectors <- SP500$GICS.Sector[SP500$Symbol %in% colnames(stock_prices)[1:r]]
stock_sectors_index <- as.numeric(as.factor(stock_sectors))
net <- graph_from_adjacency_matrix(graph_mrf$adjacency, mode = "undirected", weighted = TRUE)
colors <- c("#55efc4", "#ff7675", "#0984e3", "#a29bfe", "#B33771", "#48dbfb", "#FDA7DF", "#C4E538")
V(net)$color <- c(colors[stock_sectors_index], colors)
V(net)$type <- c(rep(FALSE, r), rep(TRUE, q))
V(net)$cluster <- c(stock_sectors_index, c(1:q))
E(net)$color <- apply(as.data.frame(get.edgelist(net)), 1,
                     function(x) ifelse(V(net)$cluster[x[1]] == V(net)$cluster[x[2]],
                                        colors[V(net)$cluster[x[1]]], 'grey'))
# where do our predictions differ from GICS?
mask <- labels_pred != stock_sectors_index
node_labels <- colnames(stock_prices)[1:r]
node_labels[!mask] <- NA
# plot network
plot(net, vertex.size = c(rep(3, r), rep(5, q)),
     vertex.label = c(node_labels, selected_sectors),
     vertex.label.cex = 0.7, vertex.label.dist = 1.0,
     vertex.frame.color = c(colors[stock_sectors_index], colors),
     layout = layout_nicely(net),
     vertex.label.family = "Helvetica", vertex.label.color = "black",
     vertex.shape = c(rep("circle", r), rep("square", q)),
     edge.width = 4*E(net)$weight)
```

## Citation
If you made use of this software please consider citing:

-   [J. V. de M. Cardoso](https://mirca.github.io), [J. Ying](https://github.com/jxying),
    [D. P. Palomar](https://www.danielppalomar.com) (2022).
    [Learning Bipartite Graphs: Heavy Tails and Multiple Components](https://papers.nips.cc/paper_files/paper/2022/hash/5adff4d5402703418f7210a4004e1314-Abstract-Conference.html).
    [Advances in Neural Information Processing Systems](https://neurips.cc/Conferences/2022) (NeurIPS’22).

## Links

- [RFinance'23 Slides](https://github.com/mirca/rfinance-talk/blob/main/rfinance.pdf)
- [NeurIPS’22 Slides](https://palomar.home.ece.ust.hk/papers/2022/CardosoYingPalomar-NeurIPS2022-slides.pdf)
- [NeurIPS'22 Poster](https://palomar.home.ece.ust.hk/papers/2022/CardosoYingPalomar-NeurIPS2022-poster.pdf)
- [NeurIPS'22 Supplementary Material](https://papers.nips.cc/paper_files/paper/2022/file/5adff4d5402703418f7210a4004e1314-Supplemental-Conference.pdf)
- [CRAN Package](https://cran.r-project.org/package=finbipartite)

Owner

GitHub Events

Issues and Pull Requests

Last synced: about 1 year ago

Packages