Struct pointcloud::PointCloud

pub struct PointCloud<M: Metric> { /* fields omitted */ }

This abstracts away data access and the distance calculation. It handles both the labels and points.

Methods

impl<M: Metric> PointCloud<M>

pub fn from_memmap_files(
    data_dim: usize,
    labels_scheme: LabelScheme,
    data_path: &[PathBuf],
    labels_path: &[PathBuf],
    ram: bool
) -> PointCloudResult<PointCloud<M>>

Builds the point cloud from a collection of memmaps. If you want it to load all the data into ram, pass a true for the ram.

pub fn from_ram(
    data: Box<[f32]>,
    data_dim: usize,
    labels: MetadataList
) -> PointCloudResult<PointCloud<M>>

Builds the point cloud from data in ram. This is for complex metadata

pub fn from_yaml(params: &Yaml) -> PointCloudResult<PointCloud<M>>

Given a yaml file on disk, it builds a point cloud. Minimal example below.

---
data_path: DATAMEMMAP
labels_path: LABELS_CSV_OR_MEMMAP
count: NUMBER_OF_DATA_POINTS
data_dim: 784
labels_dim: 10
in_ram: True

This assumes that your labels are either a CSV or a memmap file. If one specifies a schema then this is the minimal example

---
data_path: DATAMEMMAP
labels_path: LABELS_CSV_OR_MEMMAP
count: NUMBER_OF_DATA_POINTS
data_dim: 784
schema:
   natural: u32
   integer: i32
   real: f32
   string: String
   boolean: bool

pub fn from_file<P: AsRef<Path>>(path: P) -> PointCloudResult<PointCloud<M>>

Runs from_yaml on the file at a given path

pub fn simple_from_ram(
    data: Box<[f32]>,
    data_dim: usize,
    labels: Box<[f32]>,
    labels_dim: usize
) -> PointCloudResult<PointCloud<M>>

Builds the point cloud from data in ram. This is for quick things with simple metadata

pub fn len(&self) -> usize

Total number of points in the point cloud

pub fn dim(&self) -> usize

Dimension of the data in the point cloud

pub fn reference_indexes(&self) -> Vec<PointIndex>

The names of the data are currently a shallow wrapper around a usize.

pub fn get_center(&self, pn: PointIndex) -> PointCloudResult<Arc<Vec<f32>>>

Returns a arc that points to a AVX2 packed point. This also acts like a cache for these center points to ensure that we don't load multiple copies into memory. Used for heavily referenced points, like centers.

pub fn get_point(&self, pn: PointIndex) -> PointCloudResult<&[f32]>

Returns a slice corresponding to the point in question. Used for rarely referenced points, like outliers or leaves.

pub fn get_name(&self, pi: &PointIndex) -> Option<&PointName>

Gets the name from an index

pub fn get_index(&self, pn: &PointName) -> Option<&PointIndex>

Gets the index from the name

pub fn get_names(&self) -> Vec<PointName>

Gets all names in the point cloud

pub fn schema_json(&self) -> String

Gets a schema to use

pub fn get_metadata(&self, pn: PointIndex) -> PointCloudResult<Metadata>

Returns the label of a point.

This will be changed to return a label structure that can contain many different pieces of info.

pub fn get_metasummary(
    &self,
    pns: &[PointIndex]
) -> PointCloudResult<MetaSummary>

Returns a complex summary of a collection of metadatas associated to a point

pub fn distances_to_point_indices(
    &self,
    is: &[PointIndex],
    js: &[PointIndex]
) -> PointCloudResult<Vec<f32>>

The main distance function. This paralizes if there are more than 100 points.

pub fn distances_to_point_index(
    &self,
    i: PointIndex,
    indexes: &[PointIndex]
) -> PointCloudResult<Vec<f32>>

The main distance function. This paralizes if there are more than 100 points.

pub fn adj(&self, indexes: &[PointIndex]) -> PointCloudResult<AdjMatrix>

Create and adjacency matrix

pub fn distances_to_point(
    &self,
    x: &[f32],
    indexes: &[PointIndex]
) -> PointCloudResult<Vec<f32>>

The main distance function. This paralizes if there are more than 100 points.

Trait Implementations

impl<M: Metric> Debug for PointCloud<M>

fn fmt(&self, f: &mut Formatter) -> Result

Formats the value using the given formatter.