Blog – Raphael Saldanha

Handling 187 millions hospital admissions in Brazil with DuckDB

A patient geographical flow study

duckdb

sih

On ideal circumstances, any hospital admission would take place at the same city of residence of the patient. This facilitates the patient and family dislocation to the…

Persistent heat in Brazil: 1993 and 2023

climate

Persistent heat, or heat waves, are defined as sequences of days with temperatures above a reference value. These sequences of days of extreme heat are direct consequences…

Writing R packages with large datasets

package

data

This post aims to be a guide for those writing R packages that will contain datasets, presenting some approaches and solutions for some pitfalls.

Query local parquet files

database

parquet

arrow

After releasing parquet files with zonal statistics of climate indicators for Brazilian municipalities, I received some inquiries about how to query the files in an…

How 2023 was hot in different Brazilian municipalities?

climate

zonal statistics

According to the last Copernicus report, 2023 was exceptional hot year. The image bellow from this report circulated a lot at the social media.

Univariate and multivariate time series clustering

Examples with Brazilian climate data

clustering

time series

dtwclust

On this post we will try some strategies to cluster univariate and multivariate time series in R with the {dtwclust} package.

Age-adjusted COVID-19 mortality rates for Brazilian municipalities

covid19

tidyrates

On this post, we will compute crude and age-adjusted COVID-19 mortality rates for Brazilian municipalities, from 2020 to 2022 per epidemiological weeks.

Counting consecutive sequences of events: run length encoding and warm spell occurence example

rle

warm spell

sequences

Some days ago I was trying to count how many times consecutive sequences with values higher than a reference appears in a data frame.

Crude and adjusted rates in a tidy way

rates

epidemiology

tidyrates

Rates allow the comparison between the number of counts between multiple classes with different population sizes. For example, 10 disease cases that occur in 100 population…

SQLite database conversion to DuckDB and Parquet files

database

sqlite

duckdb

parquet

DuckDB is a relatively new database that works in a file, just like SQLite, but is very fast and designed for data science workflows.

Query remote parquet files with DuckDB

database

duckdb

parquet

DuckDB has a very interesting extension called httpfs that allows to query CSV and parquet files remotely, including S3 storage.

Some tips to work with SQLite database

database

sqlite

Databases are very useful for handling large-than-memory datasets, a common problem in Data Science. Several database engines work very well with R and Posit has a nice guide…