Dataframe commands (#3498)

* Sample command

* Join command with checks

* More dataframes commands

* Groupby and aggregate commands

* Missing feature dataframe flag

* Renamed file
This commit is contained in:
Fernando Herrera 2021-05-27 06:09:48 +01:00 committed by GitHub
parent d8c4b9c4fb
commit 3a5ee1aed0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
26 changed files with 1221 additions and 290 deletions

42
Cargo.lock generated
View File

@ -2843,6 +2843,12 @@ dependencies = [
"pkg-config", "pkg-config",
] ]
[[package]]
name = "libm"
version = "0.2.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c7d73b3f436185384286bd8098d17ec07c9a7d2388a6599f824d8502b529702a"
[[package]] [[package]]
name = "libnghttp2-sys" name = "libnghttp2-sys"
version = "0.1.6+1.43.0" version = "0.1.6+1.43.0"
@ -3760,9 +3766,6 @@ dependencies = [
"polars", "polars",
"serde 1.0.125", "serde 1.0.125",
"serde_bytes", "serde_bytes",
"serde_json",
"serde_yaml",
"toml",
] ]
[[package]] [[package]]
@ -4253,6 +4256,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9a64b1ec5cda2586e284722486d802acf1f7dbdc623e2bfc57e65ca1cd099290" checksum = "9a64b1ec5cda2586e284722486d802acf1f7dbdc623e2bfc57e65ca1cd099290"
dependencies = [ dependencies = [
"autocfg", "autocfg",
"libm",
] ]
[[package]] [[package]]
@ -4685,9 +4689,9 @@ dependencies = [
[[package]] [[package]]
name = "polars" name = "polars"
version = "0.13.3" version = "0.13.4"
source = "registry+https://github.com/rust-lang/crates.io-index" source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bc4e2e025126632e8e19d53cd9b655da344bd4942ba603ad246c7776b6401844" checksum = "c406ce46726b7d33b05a343d9c1317c0803a419d50bb45275de3f366410e9a80"
dependencies = [ dependencies = [
"polars-core", "polars-core",
"polars-io", "polars-io",
@ -4696,9 +4700,9 @@ dependencies = [
[[package]] [[package]]
name = "polars-arrow" name = "polars-arrow"
version = "0.13.3" version = "0.13.4"
source = "registry+https://github.com/rust-lang/crates.io-index" source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c3534c76a7bafaca9c783506a1f331ad746621d3808ab2407c02ffadd9e99326" checksum = "53b2d5fb400345c7977e4e728a10be382476f2f9d2caf6b57cd60e97ea17d364"
dependencies = [ dependencies = [
"arrow", "arrow",
"num 0.4.0", "num 0.4.0",
@ -4707,9 +4711,9 @@ dependencies = [
[[package]] [[package]]
name = "polars-core" name = "polars-core"
version = "0.13.3" version = "0.13.4"
source = "registry+https://github.com/rust-lang/crates.io-index" source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ad76c4d55017da2d0f8930b0caa327d12286c1e4407469f361e84fad176f9601" checksum = "88561e850748c507f0fc7835b35e795e770597ceecb14e0a8f7d8abf8346645d"
dependencies = [ dependencies = [
"ahash 0.7.2", "ahash 0.7.2",
"anyhow", "anyhow",
@ -4723,6 +4727,8 @@ dependencies = [
"parquet", "parquet",
"polars-arrow", "polars-arrow",
"prettytable-rs", "prettytable-rs",
"rand 0.7.3",
"rand_distr",
"rayon", "rayon",
"regex 1.5.3", "regex 1.5.3",
"thiserror", "thiserror",
@ -4731,9 +4737,9 @@ dependencies = [
[[package]] [[package]]
name = "polars-io" name = "polars-io"
version = "0.13.3" version = "0.13.4"
source = "registry+https://github.com/rust-lang/crates.io-index" source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "07f20f27363d85f847a2b7e9d1bfd426bff18680691dd42ff17ca91893f12f89" checksum = "27388810ec5f3346838725aa0aa49343802c1344b96fe82229ae781c62c98bc7"
dependencies = [ dependencies = [
"ahash 0.7.2", "ahash 0.7.2",
"anyhow", "anyhow",
@ -4755,9 +4761,9 @@ dependencies = [
[[package]] [[package]]
name = "polars-lazy" name = "polars-lazy"
version = "0.13.3" version = "0.13.4"
source = "registry+https://github.com/rust-lang/crates.io-index" source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "595906f951bacf223625ed6b0e4e73153eb9e251850bb2f9c36d78828334f32b" checksum = "0e7f83284970a9db7d0b6a56d6f944c3988587429c124c1d087188e9d2c7ad7c"
dependencies = [ dependencies = [
"ahash 0.7.2", "ahash 0.7.2",
"itertools", "itertools",
@ -5089,6 +5095,16 @@ dependencies = [
"getrandom 0.2.2", "getrandom 0.2.2",
] ]
[[package]]
name = "rand_distr"
version = "0.3.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c9e9532ada3929fb8b2e9dbe28d1e06c9b2cc65813f074fcb6bd5fbefeff9d56"
dependencies = [
"num-traits 0.2.14",
"rand 0.7.3",
]
[[package]] [[package]]
name = "rand_hc" name = "rand_hc"
version = "0.2.0" version = "0.2.0"

View File

@ -99,7 +99,10 @@ uuid_crate = { package = "uuid", version = "0.8.2", features = ["v4"], optional
which = { version = "4.1.0", optional = true } which = { version = "4.1.0", optional = true }
zip = { version = "0.5.9", optional = true } zip = { version = "0.5.9", optional = true }
polars = { version = "0.13.3",optional = true, features = ["parquet", "json"] } [dependencies.polars]
version = "0.13.4"
optional = true
features = ["parquet", "json", "random"]
[target.'cfg(unix)'.dependencies] [target.'cfg(unix)'.dependencies]
umask = "1.0.0" umask = "1.0.0"

View File

@ -188,7 +188,11 @@ pub(crate) mod touch;
pub(crate) use all::Command as All; pub(crate) use all::Command as All;
pub(crate) use any::Command as Any; pub(crate) use any::Command as Any;
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
pub(crate) use dataframe::{DataFrame, DataFrameGroupBy, DataFrameList, DataFrameLoad}; pub(crate) use dataframe::{
DataFrame, DataFrameAggregate, DataFrameConvert, DataFrameDTypes, DataFrameDrop,
DataFrameGroupBy, DataFrameJoin, DataFrameList, DataFrameLoad, DataFrameSample,
DataFrameSelect, DataFrameShow,
};
pub(crate) use enter::Enter; pub(crate) use enter::Enter;
pub(crate) use every::Every; pub(crate) use every::Every;
pub(crate) use exec::Exec; pub(crate) use exec::Exec;

View File

@ -8,6 +8,9 @@ use nu_protocol::hir::{self, Expression, ExternalRedirection, Literal, SpannedEx
use nu_protocol::{Primitive, Signature, UntaggedValue, Value}; use nu_protocol::{Primitive, Signature, UntaggedValue, Value};
use nu_table::TextStyle; use nu_table::TextStyle;
#[cfg(feature = "dataframe")]
use nu_protocol::dataframe::PolarsData;
pub struct Command; pub struct Command;
impl WholeStreamCommand for Command { impl WholeStreamCommand for Command {
@ -236,8 +239,8 @@ pub fn autoview(args: CommandArgs) -> Result<OutputStream, ShellError> {
} }
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
Value { Value {
value: UntaggedValue::DataFrame(df), value: UntaggedValue::DataFrame(PolarsData::EagerDataFrame(df)),
.. tag,
} => { } => {
if let Some(table) = table { if let Some(table) = table {
// TODO. Configure the parameter rows from file. It can be // TODO. Configure the parameter rows from file. It can be
@ -248,6 +251,20 @@ pub fn autoview(args: CommandArgs) -> Result<OutputStream, ShellError> {
let _ = result.collect::<Vec<_>>(); let _ = result.collect::<Vec<_>>();
} }
} }
#[cfg(feature = "dataframe")]
Value {
value: UntaggedValue::DataFrame(PolarsData::GroupBy(groupby)),
tag,
} => {
if let Some(table) = table {
// TODO. Configure the parameter rows from file. It can be
// adjusted to see a certain amount of values in the head
let command_args =
create_default_command_args(&context, groupby.print()?.into(), tag);
let result = table.run(command_args)?;
let _ = result.collect::<Vec<_>>();
}
}
Value { Value {
value: UntaggedValue::Primitive(Primitive::Nothing), value: UntaggedValue::Primitive(Primitive::Nothing),
.. ..

View File

@ -0,0 +1,202 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, SyntaxShape, UntaggedValue, Value,
};
use nu_source::Tagged;
use polars::frame::groupby::GroupBy;
use super::utils::convert_columns;
enum Operation {
Mean,
Sum,
Min,
Max,
First,
Last,
Nunique,
Quantile(f64),
Median,
Var,
Std,
Count,
}
impl Operation {
fn from_tagged(
name: &Tagged<String>,
quantile: Option<Tagged<f64>>,
) -> Result<Operation, ShellError> {
match name.item.as_ref() {
"mean" => Ok(Operation::Mean),
"sum" => Ok(Operation::Sum),
"min" => Ok(Operation::Min),
"max" => Ok(Operation::Max),
"first" => Ok(Operation::First),
"last" => Ok(Operation::Last),
"nunique" => Ok(Operation::Nunique),
"quantile" => {
match quantile {
None => Err(ShellError::labeled_error(
"Quantile value not fount",
"Quantile operation requires quantile value",
&name.tag,
)),
Some(value ) => {
if (value.item < 0.0) | (value.item > 1.0) {
Err(ShellError::labeled_error(
"Inappropriate quantile",
"Quantile value should be between 0.0 and 1.0",
&value.tag,
))
} else {
Ok(Operation::Quantile(value.item))
}
}
}
}
"median" => Ok(Operation::Median),
"var" => Ok(Operation::Var),
"std" => Ok(Operation::Std),
"count" => Ok(Operation::Count),
_ => Err(ShellError::labeled_error_with_secondary(
"Operation not fount",
"Operation does not exist",
&name.tag,
"Perhaps you want: mean, sum, min, max, first, last, nunique, quantile, median, count",
&name.tag,
)),
}
}
}
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls aggregate"
}
fn usage(&self) -> &str {
"Performs an aggregation operation on a groupby object"
}
fn signature(&self) -> Signature {
Signature::build("pls aggregate")
.required("operation", SyntaxShape::String, "aggregate operation")
.optional(
"selection",
SyntaxShape::Table,
"columns to perform aggregation",
)
.named(
"quantile",
SyntaxShape::Number,
"quantile value for quantile operation",
Some('q'),
)
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
aggregate(args)
}
fn examples(&self) -> Vec<Example> {
vec![Example {
description: "Aggregate sum by grouping by column a and summing on col b",
example:
"echo [[a b]; [one 1] [one 2]] | pls convert | pls groupby [a] | pls aggregate sum",
result: None,
}]
}
}
fn aggregate(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
let quantile: Option<Tagged<f64>> = args.get_flag("quantile")?;
let operation: Tagged<String> = args.req(0)?;
let op = Operation::from_tagged(&operation, quantile)?;
// Extracting the selection columns of the columns to perform the aggregation
let agg_cols: Option<Vec<Value>> = args.opt(1)?;
let (selection, agg_span) = match agg_cols {
Some(cols) => {
let (agg_string, agg_span) = convert_columns(&cols, &tag)?;
(Some(agg_string), agg_span)
}
None => (None, Span::unknown()),
};
// The operation is only done in one dataframe. Only one input is
// expected from the InputStream
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::GroupBy(nu_groupby)) = value.value {
let groupby = nu_groupby.to_groupby()?;
let groupby = match &selection {
Some(cols) => groupby.select(cols),
None => groupby,
};
let res = perform_aggregation(groupby, op, &operation.tag, &agg_span)?;
let final_df = Value {
tag,
value: UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame::new(
res,
))),
};
Ok(OutputStream::one(final_df))
} else {
Err(ShellError::labeled_error(
"No groupby in stream",
"no groupby found in input stream",
&tag,
))
}
}
}
}
fn perform_aggregation(
groupby: GroupBy,
operation: Operation,
operation_tag: &Tag,
agg_span: &Span,
) -> Result<polars::prelude::DataFrame, ShellError> {
match operation {
Operation::Mean => groupby.mean(),
Operation::Sum => groupby.sum(),
Operation::Min => groupby.min(),
Operation::Max => groupby.max(),
Operation::First => groupby.first(),
Operation::Last => groupby.last(),
Operation::Nunique => groupby.n_unique(),
Operation::Quantile(quantile) => groupby.quantile(quantile),
Operation::Median => groupby.median(),
Operation::Var => groupby.var(),
Operation::Std => groupby.std(),
Operation::Count => groupby.count(),
}
.map_err(|e| {
let span = if e.to_string().contains("Not found") {
agg_span
} else {
&operation_tag.span
};
ShellError::labeled_error("Aggregation error", format!("{}", e), span)
})
}

View File

@ -1,38 +1,26 @@
use crate::prelude::*; use crate::prelude::*;
use nu_engine::WholeStreamCommand; use nu_engine::WholeStreamCommand;
use nu_errors::ShellError; use nu_errors::ShellError;
use nu_protocol::{dataframe::NuDataFrame, Signature, UntaggedValue}; use nu_protocol::{Signature, UntaggedValue};
pub struct Command; pub struct Command;
impl WholeStreamCommand for Command { impl WholeStreamCommand for Command {
fn name(&self) -> &str { fn name(&self) -> &str {
"dataframe" "pls"
} }
fn usage(&self) -> &str { fn usage(&self) -> &str {
"Creates a dataframe from pipelined Table or List " "Commands to work with polars dataframes"
} }
fn signature(&self) -> Signature { fn signature(&self) -> Signature {
Signature::build("dataframe") Signature::build("pls")
} }
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> { fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone(); Ok(OutputStream::one(
let args = args.evaluate_once()?; UntaggedValue::string(get_full_help(&Command, args.scope())).into_value(Tag::unknown()),
))
let df = NuDataFrame::try_from_iter(args.input, &tag)?;
let init = InputStream::one(UntaggedValue::DataFrame(df).into_value(&tag));
Ok(init.to_output_stream())
}
fn examples(&self) -> Vec<Example> {
vec![Example {
description: "Takes an input stream and converts it to a dataframe",
example: "echo [[a b];[1 2] [3 4]] | dataframe",
result: None,
}]
} }
} }

View File

@ -0,0 +1,43 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, UntaggedValue,
};
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls convert"
}
fn usage(&self) -> &str {
"Converts a pipelined Table or List into a polars dataframe"
}
fn signature(&self) -> Signature {
Signature::build("pls convert")
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let args = args.evaluate_once()?;
let df = NuDataFrame::try_from_iter(args.input, &tag)?;
let init = InputStream::one(
UntaggedValue::DataFrame(PolarsData::EagerDataFrame(df)).into_value(&tag),
);
Ok(init.to_output_stream())
}
fn examples(&self) -> Vec<Example> {
vec![Example {
description: "Takes an input stream and converts it to a polars dataframe",
example: "echo [[a b];[1 2] [3 4]] | pls convert",
result: None,
}]
}
}

View File

@ -0,0 +1,97 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, SyntaxShape, UntaggedValue, Value,
};
use super::utils::convert_columns;
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls drop"
}
fn usage(&self) -> &str {
"Creates a new dataframe by dropping the selected columns"
}
fn signature(&self) -> Signature {
Signature::build("pls drop").required(
"columns",
SyntaxShape::Table,
"column names to be dropped",
)
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
drop(args)
}
fn examples(&self) -> Vec<Example> {
vec![Example {
description: "drop column a",
example: "echo [[a b]; [1 2] [3 4]] | pls convert | pls drop [a]",
result: None,
}]
}
}
fn drop(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
let columns: Vec<Value> = args.req(0)?;
let (col_string, col_span) = convert_columns(&columns, &tag)?;
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(ref df),
..
})) = value.value
{
let new_df = match col_string.iter().next() {
Some(col) => df.drop(col).map_err(|e| {
ShellError::labeled_error("Join error", format!("{}", e), &col_span)
}),
None => Err(ShellError::labeled_error(
"Empty names list",
"No column names where found",
&col_span,
)),
}?;
let res = col_string.iter().skip(1).try_fold(new_df, |new_df, col| {
new_df.drop(col).map_err(|e| {
ShellError::labeled_error("Drop error", format!("{}", e), &col_span)
})
})?;
let value = Value {
value: UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame::new(
res,
))),
tag: tag.clone(),
};
Ok(OutputStream::one(value))
} else {
Err(ShellError::labeled_error(
"No dataframe in stream",
"no dataframe found in input stream",
&tag,
))
}
}
}
}

View File

@ -0,0 +1,81 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, TaggedDictBuilder, UntaggedValue,
};
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls dtypes"
}
fn usage(&self) -> &str {
"Show dataframe data types"
}
fn signature(&self) -> Signature {
Signature::build("pls dtypes")
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
dtypes(args)
}
fn examples(&self) -> Vec<Example> {
vec![Example {
description: "drop column a",
example: "echo [[a b]; [1 2] [3 4]] | pls convert | pls dtypes",
result: None,
}]
}
}
fn dtypes(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(df),
..
})) = value.value
{
let col_names = df
.get_column_names()
.iter()
.map(|v| v.to_string())
.collect::<Vec<String>>();
let values =
df.dtypes()
.into_iter()
.zip(col_names.into_iter())
.map(move |(dtype, name)| {
let mut data = TaggedDictBuilder::new(tag.clone());
data.insert_value("column", name.as_ref());
data.insert_value("dtype", format!("{}", dtype));
data.into_value()
});
Ok(OutputStream::from_stream(values))
} else {
Err(ShellError::labeled_error(
"No dataframe in stream",
"no dataframe found in input stream",
&tag,
))
}
}
}
}

View File

@ -2,100 +2,29 @@ use crate::prelude::*;
use nu_engine::WholeStreamCommand; use nu_engine::WholeStreamCommand;
use nu_errors::ShellError; use nu_errors::ShellError;
use nu_protocol::{ use nu_protocol::{
dataframe::NuDataFrame, Primitive, Signature, SyntaxShape, UntaggedValue, Value, dataframe::{NuDataFrame, NuGroupBy, PolarsData},
Signature, SyntaxShape, UntaggedValue, Value,
}; };
use nu_source::Tagged;
use polars::frame::groupby::GroupBy;
enum Operation { use super::utils::convert_columns;
Mean,
Sum,
Min,
Max,
First,
Last,
Nunique,
Quantile(f64),
Median,
//Var,
//Std,
Count,
}
impl Operation {
fn from_tagged(
name: &Tagged<String>,
quantile: Option<Tagged<f64>>,
) -> Result<Operation, ShellError> {
match name.item.as_ref() {
"mean" => Ok(Operation::Mean),
"sum" => Ok(Operation::Sum),
"min" => Ok(Operation::Min),
"max" => Ok(Operation::Max),
"first" => Ok(Operation::First),
"last" => Ok(Operation::Last),
"nunique" => Ok(Operation::Nunique),
"quantile" => {
match quantile {
None => Err(ShellError::labeled_error(
"Quantile value not fount",
"Quantile operation requires quantile value",
&name.tag,
)),
Some(value ) => {
if (value.item < 0.0) | (value.item > 1.0) {
Err(ShellError::labeled_error(
"Inappropriate quantile",
"Quantile value should be between 0.0 and 1.0",
&value.tag,
))
} else {
Ok(Operation::Quantile(value.item))
}
}
}
}
"median" => Ok(Operation::Median),
//"var" => Ok(Operation::Var),
//"std" => Ok(Operation::Std),
"count" => Ok(Operation::Count),
_ => Err(ShellError::labeled_error_with_secondary(
"Operation not fount",
"Operation does not exist",
&name.tag,
"Perhaps you want: mean, sum, min, max, first, last, nunique, quantile, median, count",
&name.tag,
)),
}
}
}
pub struct DataFrame; pub struct DataFrame;
impl WholeStreamCommand for DataFrame { impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str { fn name(&self) -> &str {
"dataframe groupby" "pls groupby"
} }
fn usage(&self) -> &str { fn usage(&self) -> &str {
"Creates a groupby operation on a dataframe" "Creates a groupby object that can be used for other aggregations"
} }
fn signature(&self) -> Signature { fn signature(&self) -> Signature {
Signature::build("dataframe groupby") Signature::build("pls groupby").required(
.required("columns", SyntaxShape::Table, "groupby columns") "by columns",
.required( SyntaxShape::Table,
"aggregation columns", "groupby columns",
SyntaxShape::Table, )
"columns to perform aggregation",
)
.required("operation", SyntaxShape::String, "aggregate operation")
.named(
"quantile",
SyntaxShape::Number,
"auantile value for quantile operation",
Some('q'),
)
} }
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> { fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
@ -104,8 +33,8 @@ impl WholeStreamCommand for DataFrame {
fn examples(&self) -> Vec<Example> { fn examples(&self) -> Vec<Example> {
vec![Example { vec![Example {
description: "", description: "Grouping by column a",
example: "", example: "echo [[a b]; [one 1] [one 2]] | pls convert | pls groupby [a]",
result: None, result: None,
}] }]
} }
@ -115,77 +44,9 @@ fn groupby(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone(); let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?; let mut args = args.evaluate_once()?;
let quantile: Option<Tagged<f64>> = args.get_flag("quantile")?;
let operation: Tagged<String> = args.req(2)?;
let op = Operation::from_tagged(&operation, quantile)?;
// Extracting the names of the columns to perform the groupby // Extracting the names of the columns to perform the groupby
let columns: Vec<Value> = args.req(0)?; let by_columns: Vec<Value> = args.req(0)?;
let (columns_string, col_span) = convert_columns(&by_columns, &tag)?;
// Extracting the first tag from the groupby column names
let mut col_span = match columns
.iter()
.nth(0)
.map(|v| Span::new(v.tag.span.start(), v.tag.span.end()))
{
Some(span) => span,
None => {
return Err(ShellError::labeled_error(
"Empty groupby names list",
"Empty list for groupby column names",
&tag,
))
}
};
let columns_string = columns
.into_iter()
.map(|value| match value.value {
UntaggedValue::Primitive(Primitive::String(s)) => {
col_span = col_span.until(value.tag.span);
Ok(s)
}
_ => Err(ShellError::labeled_error(
"Incorrect column format",
"Only string as column name",
&value.tag,
)),
})
.collect::<Result<Vec<String>, _>>()?;
// Extracting the names of the columns to perform the aggregation
let agg_cols: Vec<Value> = args.req(1)?;
// Extracting the first tag from the aggregation column names
let mut agg_span = match agg_cols
.iter()
.nth(0)
.map(|v| Span::new(v.tag.span.start(), v.tag.span.end()))
{
Some(span) => span,
None => {
return Err(ShellError::labeled_error(
"Empty aggregation names list",
"Empty list for aggregation column names",
&tag,
))
}
};
let aggregation_string = agg_cols
.into_iter()
.map(|value| match value.value {
UntaggedValue::Primitive(Primitive::String(s)) => {
agg_span = agg_span.until(value.tag.span);
Ok(s)
}
_ => Err(ShellError::labeled_error(
"Incorrect column format",
"Only string as column name",
value.tag,
)),
})
.collect::<Result<Vec<String>, _>>()?;
// The operation is only done in one dataframe. Only one input is // The operation is only done in one dataframe. Only one input is
// expected from the InputStream // expected from the InputStream
@ -196,29 +57,31 @@ fn groupby(args: CommandArgs) -> Result<OutputStream, ShellError> {
&tag, &tag,
)), )),
Some(value) => { Some(value) => {
if let UntaggedValue::DataFrame(NuDataFrame { if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(nu_df)) = value.value {
dataframe: Some(df), let df = match nu_df.dataframe {
.. Some(df) => df,
}) = value.value None => unreachable!("No dataframe in nu_dataframe"),
{
let groupby = df
.groupby(&columns_string)
.map_err(|e| {
ShellError::labeled_error("Groupby error", format!("{}", e), col_span)
})?
.select(&aggregation_string);
let res = perform_aggregation(groupby, op, &operation.tag, &agg_span)?;
let final_df = Value {
tag,
value: UntaggedValue::DataFrame(NuDataFrame {
dataframe: Some(res),
name: "agg result".to_string(),
}),
}; };
Ok(OutputStream::one(final_df)) // This is the expensive part of the groupby; to create the
// groups that will be used for grouping the data in the
// dataframe. Once it has been done these values can be stored
// in the NuGroupBy
let groupby = df.groupby(&columns_string).map_err(|e| {
ShellError::labeled_error("Groupby error", format!("{}", e), col_span)
})?;
let groups = groupby.get_groups().to_vec();
let groupby = Value {
tag: value.tag,
value: UntaggedValue::DataFrame(PolarsData::GroupBy(NuGroupBy::new(
NuDataFrame::new_with_name(df, nu_df.name),
columns_string,
groups,
))),
};
Ok(OutputStream::one(groupby))
} else { } else {
Err(ShellError::labeled_error( Err(ShellError::labeled_error(
"No dataframe in stream", "No dataframe in stream",
@ -229,34 +92,3 @@ fn groupby(args: CommandArgs) -> Result<OutputStream, ShellError> {
} }
} }
} }
fn perform_aggregation(
groupby: GroupBy,
operation: Operation,
operation_tag: &Tag,
agg_span: &Span,
) -> Result<polars::prelude::DataFrame, ShellError> {
match operation {
Operation::Mean => groupby.mean(),
Operation::Sum => groupby.sum(),
Operation::Min => groupby.min(),
Operation::Max => groupby.max(),
Operation::First => groupby.first(),
Operation::Last => groupby.last(),
Operation::Nunique => groupby.n_unique(),
Operation::Quantile(quantile) => groupby.quantile(quantile),
Operation::Median => groupby.median(),
//Operation::Var => groupby.var(),
//Operation::Std => groupby.std(),
Operation::Count => groupby.count(),
}
.map_err(|e| {
let span = if e.to_string().contains("Not found") {
agg_span
} else {
&operation_tag.span
};
ShellError::labeled_error("Aggregation error", format!("{}", e), span)
})
}

View File

@ -0,0 +1,205 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, SyntaxShape, UntaggedValue, Value,
};
use super::utils::convert_columns;
use polars::prelude::JoinType;
use nu_source::Tagged;
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls join"
}
fn usage(&self) -> &str {
"Joins a dataframe using columns as reference"
}
fn signature(&self) -> Signature {
Signature::build("pls join")
.required("dataframe", SyntaxShape::Any, "right dataframe to join")
.required(
"l_columns",
SyntaxShape::Table,
"left column names to perform join",
)
.required(
"r_columns",
SyntaxShape::Table,
"right column names to perform join",
)
.named(
"type",
SyntaxShape::String,
"type of join. Inner by default",
Some('t'),
)
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
join(args)
}
fn examples(&self) -> Vec<Example> {
vec![
Example {
description: "inner join dataframe",
example: "echo [[a b]; [1 2] [3 4]] | pls convert | pls join $right [a] [a]",
result: None,
},
Example {
description: "right join dataframe",
example:
"echo [[a b]; [1 2] [3 4] [5 6]] | pls convert | pls join $right [b] [b] -t right",
result: None,
},
]
}
}
fn join(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
let r_df: Value = args.req(0)?;
let l_col: Vec<Value> = args.req(1)?;
let r_col: Vec<Value> = args.req(2)?;
let join_type_op: Option<Tagged<String>> = args.get_flag("type")?;
let join_type = match join_type_op {
None => JoinType::Inner,
Some(val) => match val.item.as_ref() {
"inner" => JoinType::Inner,
"outer" => JoinType::Outer,
"left" => JoinType::Left,
_ => {
return Err(ShellError::labeled_error_with_secondary(
"Incorrect join type",
"Invalid join type",
&val.tag,
"Perhaps you mean: inner, outer or left",
&val.tag,
))
}
},
};
let (l_col_string, l_col_span) = convert_columns(&l_col, &tag)?;
let (r_col_string, r_col_span) = convert_columns(&r_col, &tag)?;
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(ref df),
..
})) = value.value
{
let res = match r_df.value {
UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(r_df),
..
})) => {
// Checking the column types before performing the join
check_column_datatypes(
df,
&l_col_string,
&l_col_span,
&r_col_string,
&r_col_span,
)?;
df.join(&r_df, &l_col_string, &r_col_string, join_type)
.map_err(|e| {
ShellError::labeled_error(
"Join error",
format!("{}", e),
&l_col_span,
)
})
}
_ => Err(ShellError::labeled_error(
"Not a dataframe",
"not a dataframe type value",
&r_df.tag,
)),
}?;
let value = Value {
value: UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame::new(
res,
))),
tag: tag.clone(),
};
Ok(OutputStream::one(value))
} else {
Err(ShellError::labeled_error(
"No dataframe in stream",
"no dataframe found in input stream",
&tag,
))
}
}
}
}
fn check_column_datatypes<T: AsRef<str>>(
df: &polars::prelude::DataFrame,
l_cols: &[T],
l_col_span: &Span,
r_cols: &[T],
r_col_span: &Span,
) -> Result<(), ShellError> {
if l_cols.len() != r_cols.len() {
return Err(ShellError::labeled_error_with_secondary(
"Mismatched number of column names",
format!(
"found {} left names vs {} right names",
l_cols.len(),
r_cols.len()
),
l_col_span,
"perhaps you need to change the number of columns to join",
r_col_span,
));
}
for (l, r) in l_cols.iter().zip(r_cols.iter()) {
let l_series = df
.column(l.as_ref())
.map_err(|e| ShellError::labeled_error("Join error", format!("{}", e), l_col_span))?;
let r_series = df
.column(r.as_ref())
.map_err(|e| ShellError::labeled_error("Join error", format!("{}", e), r_col_span))?;
if l_series.dtype() != r_series.dtype() {
return Err(ShellError::labeled_error_with_secondary(
"Mismatched datatypes",
format!(
"left column type '{}' doesn't match '{}' right column match",
l_series.dtype(),
r_series.dtype()
),
l_col_span,
"perhaps you need to select other column to match",
r_col_span,
));
}
}
Ok(())
}

View File

@ -1,13 +1,16 @@
use crate::prelude::*; use crate::prelude::*;
use nu_engine::WholeStreamCommand; use nu_engine::WholeStreamCommand;
use nu_errors::ShellError; use nu_errors::ShellError;
use nu_protocol::{Signature, TaggedDictBuilder, UntaggedValue, Value}; use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, TaggedDictBuilder, UntaggedValue,
};
pub struct DataFrame; pub struct DataFrame;
impl WholeStreamCommand for DataFrame { impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str { fn name(&self) -> &str {
"dataframe list" "pls list"
} }
fn usage(&self) -> &str { fn usage(&self) -> &str {
@ -15,38 +18,46 @@ impl WholeStreamCommand for DataFrame {
} }
fn signature(&self) -> Signature { fn signature(&self) -> Signature {
Signature::build("dataframe list") Signature::build("pls list")
} }
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> { fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
let args = args.evaluate_once()?; let args = args.evaluate_once()?;
let mut dataframes: Vec<Value> = Vec::new(); let values = args
for (name, value) in args.context.scope.get_vars() { .context
if let UntaggedValue::DataFrame(df) = value.value { .scope
let mut data = TaggedDictBuilder::new(value.tag); .get_vars()
.into_iter()
.filter_map(|(name, value)| {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(df),
name: file_name,
})) = &value.value
{
let mut data = TaggedDictBuilder::new(value.tag.clone());
let polars_df = df.dataframe.unwrap(); let rows = df.height();
let cols = df.width();
let rows = polars_df.height(); data.insert_value("name", name.as_ref());
let cols = polars_df.width(); data.insert_value("file", file_name.as_ref());
data.insert_value("rows", format!("{}", rows));
data.insert_value("columns", format!("{}", cols));
data.insert_value("name", name); Some(data.into_value())
data.insert_value("file", df.name); } else {
data.insert_value("rows", format!("{}", rows)); None
data.insert_value("columns", format!("{}", cols)); }
});
dataframes.push(data.into_value()); Ok(OutputStream::from_stream(values))
}
}
Ok(OutputStream::from_stream(dataframes.into_iter()))
} }
fn examples(&self) -> Vec<Example> { fn examples(&self) -> Vec<Example> {
vec![Example { vec![Example {
description: "Lists loaded dataframes in current scope", description: "Lists loaded dataframes in current scope",
example: "dataframe list", example: "pls list",
result: None, result: None,
}] }]
} }

View File

@ -4,7 +4,8 @@ use crate::prelude::*;
use nu_engine::{EvaluatedCommandArgs, WholeStreamCommand}; use nu_engine::{EvaluatedCommandArgs, WholeStreamCommand};
use nu_errors::ShellError; use nu_errors::ShellError;
use nu_protocol::{ use nu_protocol::{
dataframe::NuDataFrame, Primitive, Signature, SyntaxShape, UntaggedValue, Value, dataframe::{NuDataFrame, PolarsData},
Primitive, Signature, SyntaxShape, UntaggedValue, Value,
}; };
use nu_source::Tagged; use nu_source::Tagged;
@ -15,7 +16,7 @@ pub struct DataFrame;
impl WholeStreamCommand for DataFrame { impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str { fn name(&self) -> &str {
"dataframe load" "pls load"
} }
fn usage(&self) -> &str { fn usage(&self) -> &str {
@ -23,7 +24,7 @@ impl WholeStreamCommand for DataFrame {
} }
fn signature(&self) -> Signature { fn signature(&self) -> Signature {
Signature::build("dataframe load") Signature::build("pls load")
.required( .required(
"file", "file",
SyntaxShape::FilePath, SyntaxShape::FilePath,
@ -67,7 +68,7 @@ impl WholeStreamCommand for DataFrame {
fn examples(&self) -> Vec<Example> { fn examples(&self) -> Vec<Example> {
vec![Example { vec![Example {
description: "Takes a file name and creates a dataframe", description: "Takes a file name and creates a dataframe",
example: "dataframe load test.csv", example: "pls load test.csv",
result: None, result: None,
}] }]
} }
@ -85,7 +86,7 @@ fn create_from_file(args: CommandArgs) -> Result<OutputStream, ShellError> {
Some("json") => from_json(args), Some("json") => from_json(args),
_ => Err(ShellError::labeled_error( _ => Err(ShellError::labeled_error(
"Error with file", "Error with file",
"Not a csv or parquet file", "Not a csv, parquet or json file",
&file.tag, &file.tag,
)), )),
}, },
@ -107,12 +108,12 @@ fn create_from_file(args: CommandArgs) -> Result<OutputStream, ShellError> {
} }
}; };
let nu_dataframe = NuDataFrame { let init = InputStream::one(
dataframe: Some(df), UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame::new_with_name(
name: file_name, df, file_name,
}; )))
.into_value(&tag),
let init = InputStream::one(UntaggedValue::DataFrame(nu_dataframe).into_value(&tag)); );
Ok(init.to_output_stream()) Ok(init.to_output_stream())
} }

View File

@ -1,9 +1,26 @@
pub mod aggregate;
pub mod command; pub mod command;
pub mod convert;
pub mod drop;
pub mod dtypes;
pub mod groupby; pub mod groupby;
pub mod join;
pub mod list; pub mod list;
pub mod load; pub mod load;
pub mod sample;
pub mod select;
pub mod show;
pub(crate) mod utils;
pub use aggregate::DataFrame as DataFrameAggregate;
pub use command::Command as DataFrame; pub use command::Command as DataFrame;
pub use convert::DataFrame as DataFrameConvert;
pub use drop::DataFrame as DataFrameDrop;
pub use dtypes::DataFrame as DataFrameDTypes;
pub use groupby::DataFrame as DataFrameGroupBy; pub use groupby::DataFrame as DataFrameGroupBy;
pub use join::DataFrame as DataFrameJoin;
pub use list::DataFrame as DataFrameList; pub use list::DataFrame as DataFrameList;
pub use load::DataFrame as DataFrameLoad; pub use load::DataFrame as DataFrameLoad;
pub use sample::DataFrame as DataFrameSample;
pub use select::DataFrame as DataFrameSelect;
pub use show::DataFrame as DataFrameShow;

View File

@ -0,0 +1,117 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, SyntaxShape, UntaggedValue, Value,
};
use nu_source::Tagged;
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls sample"
}
fn usage(&self) -> &str {
"Create sample dataframe"
}
fn signature(&self) -> Signature {
Signature::build("pls load")
.named(
"n_rows",
SyntaxShape::Number,
"number of rows to be taken from dataframe",
Some('n'),
)
.named(
"fraction",
SyntaxShape::Number,
"fraction of dataframe to be taken",
Some('f'),
)
.switch("replace", "sample with replace", Some('e'))
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
sample(args)
}
fn examples(&self) -> Vec<Example> {
vec![
Example {
description: "Sample rows from dataframe",
example: "echo [[a b]; [1 2] [3 4]] | pls load | pls sample -r 1",
result: None,
},
Example {
description: "Shows sample row using fraction and replace",
example: "echo [[a b]; [1 2] [3 4] [5 6]] | pls load | pls sample -f 0.5 -e",
result: None,
},
]
}
}
fn sample(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
let rows: Option<Tagged<usize>> = args.get_flag("n_rows")?;
let fraction: Option<Tagged<f64>> = args.get_flag("fraction")?;
let replace: bool = args.has_flag("replace");
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(ref df),
..
})) = value.value
{
let res = match (rows, fraction) {
(Some(rows), None) => df.sample_n(rows.item, replace).map_err(|e| {
ShellError::labeled_error("Polars error", format!("{}", e), &rows.tag)
}),
(None, Some(frac)) => df.sample_frac(frac.item, replace).map_err(|e| {
ShellError::labeled_error("Polars error", format!("{}", e), &frac.tag)
}),
(Some(_), Some(_)) => Err(ShellError::labeled_error(
"Incompatible flags",
"Only one selection criterion allowed",
&tag,
)),
(None, None) => Err(ShellError::labeled_error_with_secondary(
"No selection",
"No selection criterion was found",
&tag,
"Perhaps you want to use the flag -n or -f",
&tag,
)),
}?;
let value = Value {
value: UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame::new(
res,
))),
tag: tag.clone(),
};
Ok(OutputStream::one(value))
} else {
Err(ShellError::labeled_error(
"No dataframe in stream",
"no dataframe found in input stream",
&tag,
))
}
}
}
}

View File

@ -0,0 +1,84 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{
dataframe::{NuDataFrame, PolarsData},
Signature, SyntaxShape, UntaggedValue, Value,
};
use super::utils::convert_columns;
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls select"
}
fn usage(&self) -> &str {
"Creates a new dataframe with the selected columns"
}
fn signature(&self) -> Signature {
Signature::build("pls select").required(
"columns",
SyntaxShape::Table,
"selected column names",
)
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
select(args)
}
fn examples(&self) -> Vec<Example> {
vec![Example {
description: "Create new dataframe with column a",
example: "echo [[a b]; [1 2] [3 4]] | pls convert | pls select [a]",
result: None,
}]
}
}
fn select(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
let columns: Vec<Value> = args.req(0)?;
let (col_string, col_span) = convert_columns(&columns, &tag)?;
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame {
dataframe: Some(ref df),
..
})) = value.value
{
let res = df.select(&col_string).map_err(|e| {
ShellError::labeled_error("Drop error", format!("{}", e), &col_span)
})?;
let value = Value {
value: UntaggedValue::DataFrame(PolarsData::EagerDataFrame(NuDataFrame::new(
res,
))),
tag: tag.clone(),
};
Ok(OutputStream::one(value))
} else {
Err(ShellError::labeled_error(
"No dataframe in stream",
"no dataframe found in input stream",
&tag,
))
}
}
}
}

View File

@ -0,0 +1,78 @@
use crate::prelude::*;
use nu_engine::WholeStreamCommand;
use nu_errors::ShellError;
use nu_protocol::{dataframe::PolarsData, Signature, SyntaxShape, UntaggedValue};
use nu_source::Tagged;
pub struct DataFrame;
impl WholeStreamCommand for DataFrame {
fn name(&self) -> &str {
"pls show"
}
fn usage(&self) -> &str {
"Show dataframe"
}
fn signature(&self) -> Signature {
Signature::build("pls show")
.named(
"n_rows",
SyntaxShape::Number,
"number of rows to be shown",
Some('n'),
)
.switch("tail", "shows tail rows", Some('t'))
}
fn run(&self, args: CommandArgs) -> Result<OutputStream, ShellError> {
show(args)
}
fn examples(&self) -> Vec<Example> {
vec![
Example {
description: "Shows head rows from dataframe",
example: "echo [[a b]; [1 2] [3 4]] | pls convert | pls show",
result: None,
},
Example {
description: "Shows tail rows from dataframe",
example: "echo [[a b]; [1 2] [3 4] [5 6]] | pls convert | pls show -t -n 1",
result: None,
},
]
}
}
fn show(args: CommandArgs) -> Result<OutputStream, ShellError> {
let tag = args.call_info.name_tag.clone();
let mut args = args.evaluate_once()?;
let rows: Option<Tagged<usize>> = args.get_flag("rows")?;
let tail: bool = args.has_flag("tail");
match args.input.next() {
None => Err(ShellError::labeled_error(
"No input received",
"missing dataframe input from stream",
&tag,
)),
Some(value) => {
if let UntaggedValue::DataFrame(PolarsData::EagerDataFrame(df)) = value.value {
let rows = rows.map(|v| v.item);
let values = if tail { df.tail(rows)? } else { df.head(rows)? };
Ok(OutputStream::from_stream(values.into_iter()))
} else {
Err(ShellError::labeled_error(
"No dataframe in stream",
"no dataframe found in input stream",
&tag,
))
}
}
}
}

View File

@ -0,0 +1,42 @@
use crate::prelude::*;
use nu_errors::ShellError;
use nu_protocol::{Primitive, UntaggedValue, Value};
// Converts a Vec<Value> to a Vec<String> with a Span marking the whole
// location of the columns for error referencing
pub(crate) fn convert_columns<'columns>(
columns: &'columns [Value],
tag: &Tag,
) -> Result<(Vec<String>, Span), ShellError> {
let mut col_span = match columns
.iter()
.nth(0)
.map(|v| Span::new(v.tag.span.start(), v.tag.span.end()))
{
Some(span) => span,
None => {
return Err(ShellError::labeled_error(
"Empty column list",
"Empty list found for command",
tag,
))
}
};
let res = columns
.iter()
.map(|value| match &value.value {
UntaggedValue::Primitive(Primitive::String(s)) => {
col_span = col_span.until(value.tag.span);
Ok(s.clone())
}
_ => Err(ShellError::labeled_error(
"Incorrect column format",
"Only string as column name",
&value.tag,
)),
})
.collect::<Result<Vec<String>, _>>()?;
Ok((res, col_span))
}

View File

@ -253,14 +253,31 @@ pub fn create_default_context(interactive: bool) -> Result<EvaluationContext, Bo
whole_stream_command(Seq), whole_stream_command(Seq),
whole_stream_command(SeqDates), whole_stream_command(SeqDates),
whole_stream_command(TermSize), whole_stream_command(TermSize),
//Dataframe commands
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
whole_stream_command(DataFrame), whole_stream_command(DataFrame),
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
whole_stream_command(DataFrameConvert),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameLoad), whole_stream_command(DataFrameLoad),
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
whole_stream_command(DataFrameList), whole_stream_command(DataFrameList),
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
whole_stream_command(DataFrameGroupBy), whole_stream_command(DataFrameGroupBy),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameAggregate),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameShow),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameSample),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameJoin),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameDrop),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameSelect),
#[cfg(feature = "dataframe")]
whole_stream_command(DataFrameDTypes),
]); ]);
#[cfg(feature = "clipboard-cli")] #[cfg(feature = "clipboard-cli")]

View File

@ -117,7 +117,7 @@ fn helper(v: &Value) -> Result<toml::Value, ShellError> {
UntaggedValue::Error(e) => return Err(e.clone()), UntaggedValue::Error(e) => return Err(e.clone()),
UntaggedValue::Block(_) => toml::Value::String("<Block>".to_string()), UntaggedValue::Block(_) => toml::Value::String("<Block>".to_string()),
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
UntaggedValue::DataFrame(_) => toml::Value::String("<Data>".to_string()), UntaggedValue::DataFrame(_) => toml::Value::String("<DataFrame>".to_string()),
UntaggedValue::Primitive(Primitive::Range(_)) => toml::Value::String("<Range>".to_string()), UntaggedValue::Primitive(Primitive::Range(_)) => toml::Value::String("<Range>".to_string()),
UntaggedValue::Primitive(Primitive::Binary(b)) => { UntaggedValue::Primitive(Primitive::Binary(b)) => {
toml::Value::Array(b.iter().map(|x| toml::Value::Integer(*x as i64)).collect()) toml::Value::Array(b.iter().map(|x| toml::Value::Integer(*x as i64)).collect())

View File

@ -25,7 +25,9 @@ num-traits = "0.2.14"
serde = { version = "1.0", features = ["derive"] } serde = { version = "1.0", features = ["derive"] }
serde_bytes = "0.11.5" serde_bytes = "0.11.5"
polars = {version="0.13.3", optional = true} [dependencies.polars]
version = "0.13.4"
optional = true
# implement conversions # implement conversions
serde_json = "1.0" serde_json = "1.0"

View File

@ -1,3 +1,12 @@
pub mod nu_dataframe; pub mod nu_dataframe;
pub mod nu_groupby;
pub use nu_dataframe::NuDataFrame; pub use nu_dataframe::NuDataFrame;
pub use nu_groupby::NuGroupBy;
use serde::{Deserialize, Serialize};
#[derive(Debug, Clone, PartialEq, PartialOrd, Eq, Ord, Hash, Serialize, Deserialize)]
pub enum PolarsData {
EagerDataFrame(NuDataFrame),
GroupBy(NuGroupBy),
}

View File

@ -59,8 +59,18 @@ impl Default for NuDataFrame {
} }
impl NuDataFrame { impl NuDataFrame {
fn new() -> Self { pub fn new(df: polars::prelude::DataFrame) -> Self {
Self::default() NuDataFrame {
dataframe: Some(df),
name: String::from("dataframe"),
}
}
pub fn new_with_name(df: polars::prelude::DataFrame, name: String) -> Self {
NuDataFrame {
dataframe: Some(df),
name,
}
} }
} }
@ -103,7 +113,7 @@ impl<'de> Deserialize<'de> for NuDataFrame {
where where
D: Deserializer<'de>, D: Deserializer<'de>,
{ {
deserializer.deserialize_i32(NuDataFrame::new()) deserializer.deserialize_i32(NuDataFrame::default())
} }
} }
@ -137,22 +147,23 @@ impl NuDataFrame {
// Print is made out a head and if the dataframe is too large, then a tail // Print is made out a head and if the dataframe is too large, then a tail
pub fn print(&self) -> Result<Vec<Value>, ShellError> { pub fn print(&self) -> Result<Vec<Value>, ShellError> {
if let Some(df) = &self.dataframe { if let Some(df) = &self.dataframe {
let size: usize = 5; let size: usize = 20;
let mut values = self.head(Some(size))?;
if df.height() > size { if df.height() > size {
let sample_size = size / 2;
let mut values = self.head(Some(sample_size))?;
add_separator(&mut values, df); add_separator(&mut values, df);
let remaining = df.height() - sample_size;
let remaining = df.height() - size; let tail_size = remaining.min(sample_size);
let tail_size = remaining.min(size);
let mut tail_values = self.tail(Some(tail_size))?; let mut tail_values = self.tail(Some(tail_size))?;
values.append(&mut tail_values); values.append(&mut tail_values);
}
Ok(values) Ok(values)
} else {
Ok(self.head(Some(size))?)
}
} else { } else {
unreachable!() unreachable!("No dataframe found in print command")
} }
} }

View File

@ -0,0 +1,54 @@
use nu_source::Tag;
use polars::frame::groupby::{GroupBy, GroupTuples};
use serde::{Deserialize, Serialize};
use super::NuDataFrame;
use nu_errors::ShellError;
use crate::{TaggedDictBuilder, Value};
#[derive(Debug, Clone, PartialEq, PartialOrd, Eq, Ord, Hash, Serialize, Deserialize)]
pub struct NuGroupBy {
dataframe: NuDataFrame,
by: Vec<String>,
groups: GroupTuples,
}
impl NuGroupBy {
pub fn new(dataframe: NuDataFrame, by: Vec<String>, groups: GroupTuples) -> Self {
NuGroupBy {
dataframe,
by,
groups,
}
}
pub fn to_groupby(&self) -> Result<GroupBy, ShellError> {
let df = match &self.dataframe.dataframe {
Some(df) => df,
None => unreachable!("No dataframe in nu_dataframe"),
};
let by = df.select_series(&self.by).map_err(|e| {
ShellError::labeled_error("Error creating groupby", format!("{}", e), Tag::unknown())
})?;
Ok(GroupBy::new(df, by, self.groups.clone(), None))
}
pub fn print(&self) -> Result<Vec<Value>, ShellError> {
let mut values: Vec<Value> = Vec::new();
let mut data = TaggedDictBuilder::new(Tag::unknown());
data.insert_value("property", "dataframe");
data.insert_value("value", self.dataframe.name.as_ref());
values.push(data.into_value());
let mut data = TaggedDictBuilder::new(Tag::unknown());
data.insert_value("property", "group by");
data.insert_value("value", self.by.join(", "));
values.push(data.into_value());
Ok(values)
}
}

View File

@ -31,7 +31,7 @@ use std::path::PathBuf;
use std::time::SystemTime; use std::time::SystemTime;
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
use crate::dataframe::NuDataFrame; use crate::dataframe::PolarsData;
/// The core structured values that flow through a pipeline /// The core structured values that flow through a pipeline
#[derive(Debug, Clone, PartialEq, PartialOrd, Eq, Ord, Hash, Serialize, Deserialize)] #[derive(Debug, Clone, PartialEq, PartialOrd, Eq, Ord, Hash, Serialize, Deserialize)]
@ -54,7 +54,7 @@ pub enum UntaggedValue {
/// Data option that holds the polars structs required to to data /// Data option that holds the polars structs required to to data
/// manipulation and operations using polars dataframes /// manipulation and operations using polars dataframes
#[cfg(feature = "dataframe")] #[cfg(feature = "dataframe")]
DataFrame(NuDataFrame), DataFrame(PolarsData),
} }
impl UntaggedValue { impl UntaggedValue {

View File

@ -364,8 +364,8 @@ macro_rules! from_native_to_primitive {
($native_type:ty, $primitive_type:expr, $converter: expr) => { ($native_type:ty, $primitive_type:expr, $converter: expr) => {
// e.g. from u32 -> Primitive // e.g. from u32 -> Primitive
impl From<$native_type> for Primitive { impl From<$native_type> for Primitive {
fn from(int: $native_type) -> Primitive { fn from(value: $native_type) -> Primitive {
if let Some(i) = $converter(int) { if let Some(i) = $converter(value) {
$primitive_type(i) $primitive_type(i)
} else { } else {
unreachable!("Internal error: protocol did not use compatible decimal") unreachable!("Internal error: protocol did not use compatible decimal")