polars-coord-transforms 0.9.1

Tools for coordinate transforms and S2-indexing via Polars Expression Plugins

The Project

This Polars plugin provides functionality which can be loosely described as tranformation of coordinates and extraction of features from them.

It contains functions which were needed in personal and work projects, therefore its set of features might appear a bit random. Nevertheless one can find it useful in projects related to robotics, geospatial science, spatial analytics etc.

The functions are divided among three namespaces: transform, s2, distance:

• transform namespace contains functions for converting coordinates from\to map, ecef, lla, utm reference frames.

• s2 namespace contains functions which allow to work with S2 Cells

• distance namespace allows to calculate distances between coordinates.

This plugin presupposes that coordianates represent points in space and that they are expressed with struct datatype in Polars.

Getting Started

Installation

pip install polars-coord-transforms

Usage

import polars_coord_transforms

In order to use plugin, coordinates should be represented as struct with fields x, y, z (or, in case of LLA-points: lon, lat, alt)!

For instance, if coordinates are in separate columns, one can make a valid struct with pl.struct native Polars function:

import polars as pl

df = pl.DataFrame(
    dict(
            lon=[31.409197919000064,],
            lat=[58.860667429000046,],
            alt=[57.309668855211015,],
        )
)

df.with_columns(
    point=pl.struct("lon", "lat", "alt")
)

Examples

Suppose we have the following DataFrame with some coordinates (column "pose"), rotation quaternion (column "rotation") and offset vector (column "offset"):

import polars as pl

df = pl.DataFrame(
    [
        pl.Series("pose", [{'x': 4190.66735544079, 'y': 14338.862844330957, 'z': 10.96391354687512}], dtype=pl.Struct({'x': pl.Float64, 'y': pl.Float64, 'z': pl.Float64})),
        pl.Series("rotation", [{'x': 0.13007119, 'y': 0.26472049, 'z': 0.85758219, 'w': 0.42137553}], dtype=pl.Struct({'x': pl.Float64, 'y': pl.Float64, 'z': pl.Float64, 'w': pl.Float64})),
        pl.Series("offset", [{'x': 2852423.40536658, 'y': 2201848.41975346, 'z': 5245234.74365368}], dtype=pl.Struct({'x': pl.Float64, 'y': pl.Float64, 'z': pl.Float64})),
    ]
)
print(df)


shape: (1, 3)
┌─────────────────────────────┬───────────────────────────┬───────────────────────────────────┐
│ pose                        ┆ rotation                  ┆ offset                            │
│ ---                         ┆ ---                       ┆ ---                               │
│ struct[3]                   ┆ struct[4]                 ┆ struct[3]                         │
╞═════════════════════════════╪═══════════════════════════╪═══════════════════════════════════╡
│ {4190.667,14338.863,10.964} ┆ {0.130,0.265,0.858,0.421} ┆ {2852423.405,2201848.420,5245234… │
└─────────────────────────────┴───────────────────────────┴───────────────────────────────────┘

transform

Transform coordinates from map reference frame to ECEF (Earth-Ceneterd, Earth-Fixed) coordinate system using a rotation quaternion and an offset vector.

df.with_columns(
    ecef=pl.col("pose").transform.map_to_ecef(
        pl.col("rotation"), pl.col("offset")
    )
)


shape: (1, 4)
┌────────────────────────┬────────────────────────┬────────────────────────┬───────────────────────┐
│ pose                   ┆ rotation               ┆ offset                 ┆ ecef                  │
│ ---                    ┆ ---                    ┆ ---                    ┆ ---                   │
│ struct[3]              ┆ struct[4]              ┆ struct[3]              ┆ struct[3]             │
╞════════════════════════╪════════════════════════╪════════════════════════╪═══════════════════════╡
│ {4190.667,14338.863,10 ┆ {0.130,0.265,0.858,0.4 ┆ {2852423.405,2201848.4 ┆ {2840491.941,2197932. │
│ .964}                  ┆ 21}                    ┆ 20,5245234…            ┆ 225,5253325…          │
└────────────────────────┴────────────────────────┴────────────────────────┴───────────────────────┘

Inverse transformation from ECEF to map

df.with_columns(
    pose_new=pl.col("ecef").transform.ecef_to_map("rotation", "offset")
).select(
    "pose",
    "pose_new"
)


shape: (1, 5)
┌───────────────────┬───────────────────┬───────────────────┬───────────────────┬──────────────────┐
│ pose              ┆ rotation          ┆ offset            ┆ ecef              ┆ pose_new         │
│ ---               ┆ ---               ┆ ---               ┆ ---               ┆ ---              │
│ struct[3]         ┆ struct[4]         ┆ struct[3]         ┆ struct[3]         ┆ struct[3]        │
╞═══════════════════╪═══════════════════╪═══════════════════╪═══════════════════╪══════════════════╡
│ {4190.667,14338.8 ┆ {0.130,0.265,0.85 ┆ {2852423.405,2201 ┆ {2840491.941,2197 ┆ {4190.667,14338. │
│ 63,10.964}        ┆ 8,0.421}          ┆ 848.420,5245234…  ┆ 932.225,5253325…  ┆ 863,10.964}      │
└───────────────────┴───────────────────┴───────────────────┴───────────────────┴──────────────────┘

Transform coordinates from ECEF to LLA (Longitude, Latitude, Altitude)

df.with_columns(
    lla=pl.col("ecef").transform.ecef_to_lla()
)

shape: (1, 3)
┌─────────────────────────────┬───────────────────────────────────┬─────────────────────────┐
│ pose                        ┆ ecef                              ┆ lla                     │
│ ---                         ┆ ---                               ┆ ---                     │
│ struct[3]                   ┆ struct[3]                         ┆ struct[3]               │
╞═════════════════════════════╪═══════════════════════════════════╪═════════════════════════╡
│ {4190.667,14338.863,10.964} ┆ {2840491.941,2197932.225,5253325… ┆ {37.732,55.820,163.916} │
└─────────────────────────────┴───────────────────────────────────┴─────────────────────────┘

Inverse transform from LLA to ECEF

df.with_columns(
    ecef_new=pl.col("lla").transform.lla_to_ecef()
)


shape: (1, 4)
┌────────────────────────┬────────────────────────┬────────────────────────┬───────────────────────┐
│ pose                   ┆ ecef                   ┆ lla                    ┆ ecef_new              │
│ ---                    ┆ ---                    ┆ ---                    ┆ ---                   │
│ struct[3]              ┆ struct[3]              ┆ struct[3]              ┆ struct[3]             │
╞════════════════════════╪════════════════════════╪════════════════════════╪═══════════════════════╡
│ {4190.667,14338.863,10 ┆ {2840491.941,2197932.2 ┆ {37.732,55.820,163.916 ┆ {2840491.941,2197932. │
│ .964}                  ┆ 25,5253325…            ┆ }                      ┆ 225,5253325…          │
└────────────────────────┴────────────────────────┴────────────────────────┴───────────────────────┘

Transform coordinates from LLA to UTM coordinates (UTM zone is derived from coordinates themselves)

df.with_columns(
    utm=pl.col("lla").transform.lla_to_utm()
)


shape: (1, 3)
┌─────────────────────────────┬─────────────────────────┬──────────────────────────────────┐
│ pose                        ┆ lla                     ┆ utm                              │
│ ---                         ┆ ---                     ┆ ---                              │
│ struct[3]                   ┆ struct[3]               ┆ struct[3]                        │
╞═════════════════════════════╪═════════════════════════╪══════════════════════════════════╡
│ {4190.667,14338.863,10.964} ┆ {37.732,55.820,163.916} ┆ {420564.380,6186739.936,163.916} │
└─────────────────────────────┴─────────────────────────┴──────────────────────────────────┘

Find UTM zone number from a LLA point

df.with_columns(
    utm_zone_number=pl.col("lla").transform.lla_to_utm_zone_number()
)

shape: (1, 3)
┌─────────────────────────┬──────────────────────────────────┬─────────────────┐
│ lla                     ┆ utm                              ┆ utm_zone_number │
│ ---                     ┆ ---                              ┆ ---             │
│ struct[3]               ┆ struct[3]                        ┆ u8              │
╞═════════════════════════╪══════════════════════════════════╪═════════════════╡
│ {37.732,55.820,163.916} ┆ {420564.380,6186739.936,163.916} ┆ 37              │
└─────────────────────────┴──────────────────────────────────┴─────────────────┘

s2

Find S2 CellID of a point with longitude and latitude (with a given cell level)

df.select(
    cellid_30=pl.col("lla").s2.lonlat_to_cellid(level=30),
    cellid_28=pl.col("lla").s2.lonlat_to_cellid(level=28),
    cellid_5=pl.col("lla").s2.lonlat_to_cellid(level=5),
)


shape: (1, 3)
┌─────────────────────┬─────────────────────┬─────────────────────┐
│ cellid_30           ┆ cellid_28           ┆ cellid_5            │
│ ---                 ┆ ---                 ┆ ---                 │
│ u64                 ┆ u64                 ┆ u64                 │
╞═════════════════════╪═════════════════════╪═════════════════════╡
│ 5095036114269810839 ┆ 5095036114269810832 ┆ 5094697078462873600 │
└─────────────────────┴─────────────────────┴─────────────────────┘

Find longitude and latitude from a S2 CellID

df.select(
    lla_cell=pl.lit(5095036114269810839, dtype=pl.UInt64()).s2.cellid_to_lonlat()
)

shape: (1, 1)
┌─────────────────┐
│ lla_cell        │
│ ---             │
│ struct[2]       │
╞═════════════════╡
│ {37.732,55.820} │
└─────────────────┘

Find whether a given LLA point is in a S2 Cell identified by a specific ID

df.select(
    lla",
    cellid=pl.lit(5095036114269810832, dtype=pl.UInt64()),
    is_in_cell=pl.lit(5095036114269810832, dtype=pl.UInt64()).s2.cell_contains_point(pl.col("lla"))
)


shape: (1, 3)
┌─────────────────────────┬─────────────────────┬────────────┐
│ lla                     ┆ cellid              ┆ is_in_cell │
│ ---                     ┆ ---                 ┆ ---        │
│ struct[3]               ┆ u64                 ┆ bool       │
╞═════════════════════════╪═════════════════════╪════════════╡
│ {37.732,55.820,163.916} ┆ 5095036114269810832 ┆ true       │
└─────────────────────────┴─────────────────────┴────────────┘

Find vertices of a S2 Cell from a CellID

df.with_columns(
    cellid=pl.col("lla").s2.lonlat_to_cellid(level=5),
).with_columns(
    vertices=pl.col("cellid").s2.cellid_to_vertices()
)

shape: (1, 4)
┌─────────────────────────┬─────────────────────────┬─────────────────────┬────────────────────────┐
│ pose                    ┆ lla                     ┆ cellid              ┆ vertices               │
│ ---                     ┆ ---                     ┆ ---                 ┆ ---                    │
│ struct[3]               ┆ struct[3]               ┆ u64                 ┆ struct[8]              │
╞═════════════════════════╪═════════════════════════╪═════════════════════╪════════════════════════╡
│ {4190.667,14338.863,10. ┆ {37.732,55.820,163.916} ┆ 5094697078462873600 ┆ {37.304,55.491,40.932, │
│ 964}                    ┆                         ┆                     ┆ 57.545,36.…            │
└─────────────────────────┴─────────────────────────┴─────────────────────┴────────────────────────┘

df.select("vertices").unnest("vertices")

shape: (1, 8)
┌────────┬────────┬────────┬────────┬────────┬────────┬────────┬────────┐
│ v0_lon ┆ v0_lat ┆ v1_lon ┆ v1_lat ┆ v2_lon ┆ v2_lat ┆ v3_lon ┆ v3_lat │
│ ---    ┆ ---    ┆ ---    ┆ ---    ┆ ---    ┆ ---    ┆ ---    ┆ ---    │
│ f64    ┆ f64    ┆ f64    ┆ f64    ┆ f64    ┆ f64    ┆ f64    ┆ f64    │
╞════════╪════════╪════════╪════════╪════════╪════════╪════════╪════════╡
│ 37.304 ┆ 55.491 ┆ 40.932 ┆ 57.545 ┆ 36.495 ┆ 59.135 ┆ 33.024 ┆ 56.886 │
└────────┴────────┴────────┴────────┴────────┴────────┴────────┴────────┘

distance

df = pl.DataFrame(
    [
        pl.Series("point_1", [{'x': -8893.663914126577, 'y': 19116.178523519542, 'z': 14.98697863612324}], dtype=pl.Struct({'x': pl.Float64, 'y': pl.Float64, 'z': pl.Float64})),
        pl.Series("point_2", [{'x': 1553.3742543335538, 'y': 2916.118342842441, 'z': 15.580027717165649}], dtype=pl.Struct({'x': pl.Float64, 'y': pl.Float64, 'z': pl.Float64})),
    ]
)

Find Euclidean distance between two points using all 3 components of a point-vector

df.with_columns(
    distance=pl.col("point_1").distance.euclidean_3d(pl.col("point_2"))
)

shape: (1, 3)
┌──────────────────────────────┬────────────────────────────┬───────────┐
│ point_1                      ┆ point_2                    ┆ distance  │
│ ---                          ┆ ---                        ┆ ---       │
│ struct[3]                    ┆ struct[3]                  ┆ f64       │
╞══════════════════════════════╪════════════════════════════╪═══════════╡
│ {-8893.664,19116.179,14.987} ┆ {1553.374,2916.118,15.580} ┆ 19276.477 │
└──────────────────────────────┴────────────────────────────┴───────────┘

Find cosine similarity between between two points using all 3 components of a point-vector

df.with_columns(
    cosine_sim=pl.col("point_1").distance.cosine_similarity_3d(pl.col("point_2"))
)

shape: (1, 3)
┌──────────────────────────────┬────────────────────────────┬────────────┐
│ point_1                      ┆ point_2                    ┆ cosine_sim │
│ ---                          ┆ ---                        ┆ ---        │
│ struct[3]                    ┆ struct[3]                  ┆ f64        │
╞══════════════════════════════╪════════════════════════════╪════════════╡
│ {-8893.664,19116.179,14.987} ┆ {1553.374,2916.118,15.580} ┆ 0.602      │
└──────────────────────────────┴────────────────────────────┴────────────┘

Find Euclidean distance between two points using 2 components of a point-vector (X and Y)

df.with_columns(
    distance=pl.col("point_1").distance.euclidean_2d(pl.col("point_2"))
)

┌──────────────────────────────┬────────────────────────────┬───────────┐
│ point_1                      ┆ point_2                    ┆ distance  │
│ ---                          ┆ ---                        ┆ ---       │
│ struct[3]                    ┆ struct[3]                  ┆ f64       │
╞══════════════════════════════╪════════════════════════════╪═══════════╡
│ {-8893.664,19116.179,14.987} ┆ {1553.374,2916.118,15.580} ┆ 19276.477 │
└──────────────────────────────┴────────────────────────────┴───────────┘

Find cosine similarity between between two points using 2 components of a point-vector (X and Y)

shape: (1, 3)
┌──────────────────────────────┬────────────────────────────┬────────────┐
│ point_1                      ┆ point_2                    ┆ cosine_sim │
│ ---                          ┆ ---                        ┆ ---        │
│ struct[3]                    ┆ struct[3]                  ┆ f64        │
╞══════════════════════════════╪════════════════════════════╪════════════╡
│ {-8893.664,19116.179,14.987} ┆ {1553.374,2916.118,15.580} ┆ 0.602      │
└──────────────────────────────┴────────────────────────────┴────────────┘