vidyasagarv.com
Python bindings to oxyroot. Makes reading .root files blazing fast 🚀
arrays to return a polars dataframe
+4
-2
Cargo.toml
+4
-2
Cargo.toml
+8
-6
README.md
+8
-6
README.md
···
8
9
A fast, Rust-powered Python reader for CERN ROOT files.
10
11
-
This package provides a simple and Pythonic interface bindings to `oxyroot`, a rust package, to read data from `.root` files, inspired by libraries like `uproot`. It leverages the speed of Rust for high-performance data extraction and integrates with the scientific Python ecosystem by providing data as NumPy arrays.
12
13
## Features
14
15
-
- **High-Performance**: Core logic is written in Rust for maximum speed.
16
-
- **Parquet Conversion**: Convert TTrees directly to Apache Parquet files with a single command.
17
-
- **NumPy Integration**: Get branch data directly as NumPy arrays.
18
-
- **Simple, Pythonic API**: Easy to learn and use, and similar to `uproot`
19
20
## Quick Start
21
···
60
61
## Performance
62
63
-
`oxyroot` is designed to be fast. Here is a simple benchmark comparing the time taken to read all branches of a TTree with `uproot` and `oxyroot`.
64
65
```python
66
import oxyroot
···
87
end_time = time.time()
88
print(f"Oxyroot took: {end_time - start_time:.3f}s")
89
```
90
91
## License
92
···
8
9
A fast, Rust-powered Python reader for CERN ROOT files.
10
11
+
This python package provides simple bindings to [`oxyroot`, a rust package](https://github.com/m-dupont/oxyroot), to read data from `.root` files, inspired by libraries like `uproot`. It leverages the speed of Rust and integrates with the scientific Python ecosystem by providing data as NumPy arrays or polars dataframes.
12
13
## Features
14
15
+
- Simple API similar to `uproot`
16
+
- Core logic is written in Rust.
17
+
- Get branch data directly as NumPy arrays or Polars dataframe.
18
+
- Parquet Conversion: Convert TTrees directly to Apache Parquet files with a single command.
19
20
## Quick Start
21
···
60
61
## Performance
62
63
+
`oxyroot` is intended to be fast. Here is a simple benchmark comparing the time taken to read all branches of a TTree with `uproot` and `oxyroot`.
64
65
```python
66
import oxyroot
···
87
end_time = time.time()
88
print(f"Oxyroot took: {end_time - start_time:.3f}s")
89
```
90
+
91
+
On a small file (~20 MB) with multiple data formats, oxyroot took half the time of uproot, and also read in the branch with strings!
92
93
## License
94
+69
src/lib.rs
+69
src/lib.rs
···
14
use parquet::arrow::ArrowWriter;
15
use parquet::basic::{BrotliLevel, Compression, GzipLevel, ZstdLevel};
16
use parquet::file::properties::WriterProperties;
17
18
#[pyclass(name = "RootFile")]
19
struct PyRootFile {
···
92
branches: branches.into_iter(),
93
},
94
)
95
}
96
97
#[pyo3(signature = (output_file, overwrite = false, compression = "snappy", columns = None))]
···
14
use parquet::arrow::ArrowWriter;
15
use parquet::basic::{BrotliLevel, Compression, GzipLevel, ZstdLevel};
16
use parquet::file::properties::WriterProperties;
17
+
use polars::prelude::*;
18
+
use pyo3_polars::PyDataFrame;
19
20
#[pyclass(name = "RootFile")]
21
struct PyRootFile {
···
94
branches: branches.into_iter(),
95
},
96
)
97
+
}
98
+
99
+
#[pyo3(signature = (columns = None))]
100
+
fn arrays(&self, columns: Option<Vec<String>>) -> PyResult<PyDataFrame> {
101
+
let mut file =
102
+
RootFile::open(&self.path).map_err(|e| PyValueError::new_err(e.to_string()))?;
103
+
let tree = file
104
+
.get_tree(&self.name)
105
+
.map_err(|e| PyValueError::new_err(e.to_string()))?;
106
+
107
+
let branches_to_save = if let Some(columns) = columns {
108
+
columns
109
+
} else {
110
+
tree.branches().map(|b| b.name().to_string()).collect()
111
+
};
112
+
113
+
let mut series_vec = Vec::new();
114
+
115
+
for branch_name in branches_to_save {
116
+
let branch = match tree.branch(&branch_name) {
117
+
Some(branch) => branch,
118
+
None => {
119
+
println!("Branch '{}' not found, skipping", branch_name);
120
+
continue;
121
+
}
122
+
};
123
+
124
+
let series = match branch.item_type_name().as_str() {
125
+
"float" => {
126
+
let data = branch.as_iter::<f32>().unwrap().collect::<Vec<_>>();
127
+
Series::new((&branch_name).into(), data)
128
+
}
129
+
"double" => {
130
+
let data = branch.as_iter::<f64>().unwrap().collect::<Vec<_>>();
131
+
Series::new((&branch_name).into(), data)
132
+
}
133
+
"int32_t" => {
134
+
let data = branch.as_iter::<i32>().unwrap().collect::<Vec<_>>();
135
+
Series::new((&branch_name).into(), data)
136
+
}
137
+
"int64_t" => {
138
+
let data = branch.as_iter::<i64>().unwrap().collect::<Vec<_>>();
139
+
Series::new((&branch_name).into(), data)
140
+
}
141
+
"uint32_t" => {
142
+
let data = branch.as_iter::<u32>().unwrap().collect::<Vec<_>>();
143
+
Series::new((&branch_name).into(), data)
144
+
}
145
+
"uint64_t" => {
146
+
let data = branch.as_iter::<u64>().unwrap().collect::<Vec<_>>();
147
+
Series::new((&branch_name).into(), data)
148
+
}
149
+
"string" => {
150
+
let data = branch.as_iter::<String>().unwrap().collect::<Vec<_>>();
151
+
Series::new((&branch_name).into(), data)
152
+
}
153
+
other => {
154
+
println!("Unsupported branch type: {}, skipping", other);
155
+
continue;
156
+
}
157
+
};
158
+
series_vec.push(series);
159
+
}
160
+
161
+
let df = DataFrame::new(series_vec.into_iter().map(|s| s.into()).collect())
162
+
.map_err(|e| PyValueError::new_err(e.to_string()))?;
163
+
Ok(PyDataFrame(df))
164
}
165
166
#[pyo3(signature = (output_file, overwrite = false, compression = "snappy", columns = None))]