analysis.Rmd

---
title: "Real State Exploration"
author: "Gabriel Bessa"
date: "August 17, 2018"
output:
  html_document:
    df_print: paged
---

# Introduction

Moving away from home for the first time is often a experience that most of people will always remember. This experience can be either really good or really bad. I know that someday it will be my turn, so I decided to prepare my self to do it, and the first step will be understanding how much would cost to rent an apartment in my hometown.

This report will explore a dataset containing rental prices and attributes for approximately 2218 apartments that are located at Belo Horizonte, Brazil. This data set was crawled from [NetImoveis](https://www.netimoveis.com/) using the [NetSpider](https://github.com/gabrielvcbessa/real-state-crawlers/blob/master/net_spider.py) crawler.

```{r echo=FALSE, message=FALSE, warning=FALSE}
df <- read.csv('net_apartments.csv')
```

# Analysis
## Plotting locations

When handling a dataset that has locations on each observation, the first thing that we might want to do is plotting each observation in a map, and this is what we are going to do.

```{r echo=FALSE, message=FALSE, warning=FALSE}
library(ggmap)
library(dplyr)
library(maps)

invisible(gmaps.key <- scan('maps.key', what = 'string'))
invisible(gmaps.style <- scan('ultra_light.style', what = 'string'))

map.apartments <- get_googlemap(
  center = c(lon = mean(df$longitude), lat = mean(df$latitude)),
  maptype = 'roadmap',
  zoom = 3,
  style = gmaps.style,
  key = gmaps.key)

ggmap(map.apartments) +
  geom_point(aes(x = longitude, y = latitude), 
             data = df, alpha = .75) +
  theme(plot.margin = grid::unit(c(0, 0, 0, 0), 'mm')) 

```

The plot show us that we have some misplaced points. Some of them have 0 value for the latitude and longitude, and others are placed on different brazilian capitals. We are going to impute those values and generate our map again.

```{r echo=FALSE, message=FALSE, warning=FALSE}
library(geojsonio)
library(gridExtra)
library(rgdal)

outlier_values <- boxplot.stats(df$longitude)$out
outlier_indexes <- which(df$longitude %in% outlier_values)
outliers <- df[outlier_indexes, ]

# We are going to input some values to the missing locations
df.by_neigh <- df[-outlier_indexes, ] %>% 
  group_by(neighborhood) %>% 
  summarise(latitude = mean(latitude),
            longitude = mean(longitude))

outliers.imputed <- 
  transform(outliers,
            latitude = df.by_neigh[neighborhood == neighborhood, ]$latitude,
            longitude = df.by_neigh[neighborhood == neighborhood, ]$longitude)

df.imputed <- df
df.imputed[outlier_indexes, ] <- outliers.imputed

bh.json <- geojson_read('bh.geojson', what = 'sp')
bh.poly <- fortify(bh.json)

bh.map <- get_googlemap(
  center = c(lon = mean(bh.poly$long), lat = mean(bh.poly$lat)),
  maptype = 'roadmap',
  zoom = 11,
  style = gmaps.style,
  key = gmaps.key)

bh.bbox <- attr(bh.map, 'bb')

ggmap(bh.map, padding = 0) +
  geom_polygon(data = bh.poly, 
               aes(long, lat),
               size = .55,
               colour = '#6692CC',
               fill = NA,
               alpha = 0.1) +
  stat_density2d(data = df.imputed, 
                 aes(longitude, latitude, 
                     fill = ..level.., 
                     alpha = ..level..), 
                 size = 1, bins = 50, geom = 'polygon') +
  scale_fill_gradient('Real estate\nDensity', low = 'blue', high = 'orange') +
  scale_alpha(range = c(.1, .3), guide = FALSE)
```

With this heatmap we can see that we have observations from most neighborhoods of the city, but in the south-west region we have a higher density.