我想阅读以下(相当破碎,IMO)来自Rigol MSO 5000示波器的带有Pola-rs的CSV:
D7-D0,D15-D8,t0 = -25.01s, tInc = 2e-06,
+2.470000E+02,2.000000E+00,,
+1.590000E+02,1.600000E+01,,
+2.400000E+02,2.000000E+00,,
+2.470000E+02,+1.300000E+02,,
+1.590000E+02,1.800000E+01,,
+2.470000E+02,+1.300000E+02,,
9.500000E+01,1.800000E+01,,
9.500000E+01,1.800000E+01,,
+2.400000E+02,0.000000E+00,,
(...)
下面是我目前的Jupyter笔记本迭代/尝试之前,发现选择与索引是不鼓励在Pola-rs:
import polars as pl
df = pl.read_csv("normal0.csv")
# Grab initial condition and increments
t0 = df.columns[2]; assert "t0" in t0; t0 = float(t0.split('=')[1].replace('s', '').strip())
tinc = df.columns[3]; assert "tInc" in tinc; tinc = float(tinc.split('=')[1].strip())
# TODO: Generate Series() from t0+tinc and include it in the final DataFrame
# Reshape and cleanup
probes = df.with_column(df["D7-D0"].cast(pl.Float32).alias("D0-D7 floats")) \
.with_column(df["D15-D8"].cast(pl.Float32).alias("D15-D8 floats")) \
.drop(df.columns[2]) \
.drop(df.columns[3])
(...)
def split_probes(probes: pl.Series):
''' Splits bundles of probe cables such as D0-D7 or D15-D8 into individual dataframe columns
'''
out = pl.DataFrame(columns=["D"+str(line) for line in range(0,16)])
# for row in range(probes.height):
# for probe in range(0, 7):
# out["D"+str(probe)].with_columns(probes["D0-D7 floats"][row % (probe + 1)])
# for probe in reversed(range(9, 16)): # TODO: Fix future captures or parametrise this
# outprobes["D15-D8 floats"][row % probe]
下面是我在I was told on Polars Discord that this problem might not be optimally solvable with dataframe libraries时的较低级别CSV解析Rust伪代码方法:
use csv;
use std::error::Error;
use std::io;
use std::process;
use serde::Deserialize;
#[derive(Debug, Deserialize)]
struct OrigOscilloscope {
#[serde(rename = "D7-D0")]
d7_d0: String, // TODO: Unfortunately those fields are "user-flippable" in order from the scope, i.e: d0_d7 vs d7_d0
#[serde(rename = "D15-D8")]
d15_d8: String,
// Do not even register those on Serde as they are empty rows anyway
// t0: Option<String>,
// t_inc: Option<String>
}
#[derive(Debug, Deserialize)]
struct LAProbesOscilloscopeState {
//Vec<d0...d15>: Vec<Vec<16*f32>>, // TODO: More appropriate struct representation for the target dataframe
d0: f32,
d1: f32,
d2: f32,
d3: f32,
d4: f32,
d5: f32,
d6: f32,
d7: f32,
d8: f32,
d9: f32,
d10: f32,
d11: f32,
d12: f32,
d13: f32,
d14: f32,
d15: f32,
timestamp: f32
}
fn run() -> Result<(), Box<dyn Error>> {
let mut rdr = csv::ReaderBuilder::new()
.has_headers(false)
.flexible(true) // ignore broken header
.from_reader(io::stdin());
// Get timeseries information from header... :facepalm: rigol
// Initial timestamp...
let header = rdr.headers()?.clone();
let t0_header: Vec<&str> = header[2].split('=').collect();
let t0 = t0_header[1].trim_start().replace('s', "").parse::<f32>()?;
// ...and increments
let tinc_header: Vec<&str> = header[3].split('=').collect();
let tinc = tinc_header[1].trim_start().parse::<f32>()?;
println!("Initial timestamp {t0} with increments of {tinc} seconds");
// Now do the splitting of wires from its Dx-Dy "bundles"
let mut timestamp = t0;
for result in rdr.deserialize().skip(1) {
let row: OrigOscilloscope = result?;
// if rdr.position().line().rem_euclid(8) {
// // Read D0-D15 field(s), expanding them into the current row, matching its column
// }
// println!("{:#?}", row.d7_d0.parse::<f32>()?);
// update timestamp for this row
timestamp = timestamp + tinc;
}
Ok(())
}
fn main() {
if let Err(err) = run() {
println!("{}", err);
process::exit(1);
}
}
我希望可以清楚地看到所需的目标 Dataframe 如下所示:
d0, d1, d2, d3, d4, d5, d6, d7, d8, d9, d10, d11, d12, d13, d14, d15, timestamp
95, 95, 247, 159, 247, 240, 159, 247 (...) -25.01+000000.2
(...)
并且生成的代码应该(理想情况下?)高效地使用Polars。另外,如上所述,避免“使用索引进行选择”,因为它可能在未来的Polars中完全被弃用。
1条答案
按热度按时间igetnqfo1#
也许您正在寻找
unstack
(它的性能令人难以置信)。让我们从这个数据开始(它只是复制您提供的值),我还更改了列名,以便更容易检查:
由于我不熟悉您的数据源,我将假设您的输入数据具有规则化的模式--特别是始终提供D7-D0(即csv文件中没有跳过的行)。
如果是这样的话,这里有一些高性能的代码,应该得到球滚动...
一个二个一个一个
它没有你想要的列名(你可以改变),也不包含时间信息,但是它可以帮助你朝着正确的方向开始。