# Specifying a filter
The `vertexFilter` is used to specify filters on a per-label basis for vertices. This allows you to control which vertex labels and properties are included in the export.
+ `vertexFilter` - This is the top-level field for specifying vertex filters.
+ If the `vertexFilter` is not provided at all, then all vertex properties for all vertex labels will be exported. If the `vertexFilter` is provided but is an empty object, then no vertices will be exported.
+ Each key in the `vertexFilter` object corresponds to a vertex label that you want to describe a filter for. For example, `"Person"` or `"Organization"`.
+ For each vertex label key, the value is an object with a `"properties"` field.
+ The "properties" field allows you to specify which properties of that vertex label should be included in the export. Each property is defined by a key-value pair, where the key is the desired output property name (e.g. `"name"`), and the value is an object with the following fields:
+ `outputType`: Specifies the data type to use for the property in the exported data (e.g. "String", "Int", "Float"). For a full-list of supported types and the corresponding type names that can be used in filtering, see [Using CSV data](using-CSV-data.md). If a type is not provided, the export process will determine the type. If a given property is present as multiple types (e.g. one vertex has `"height"` stored as a double, and another edge has it stored as a string), the type will be of `Any` type. Otherwise, it will be the type of the property as present in vertices.
+ `sourcePropertyName`: The name of the property as it exists in the original graph data. If not provided, it is assumed that the key matches the desired `sourcePropertyName`.
+ `multiValueHandling`: Specifies how to handle properties that have multiple values. Can be either "TO\_LIST" to export all values as a list, or "PICK\_FIRST" to export the first value encountered. If not specified, the default value is "PICK\_FIRST".
`edgeFilter` is used to specify filters on a per-label basis for edges. This allows you to control which edge labels and properties are included in the export.
+ `edgeFilter` - This is the top-level field for specifying edge filters.
+ If the `edgeFilter` is not provided at all, then all edge properties for all edge labels will be exported. If the `edgeFilter` is provided but is an empty object, then no edges will be exported.
+ Each key in the `edgeFilter` object corresponds to a edge label that you want to describe a filter for. For example, `"knows"` or `"friendOf"`.
+ For each edge label key, the value is an object with a `"properties"` field.
+ The "properties" field allows you to specify which properties of that edge label should be included in the export. Each property is defined by a key-value pair, where the key is the desired output property name (e.g. `"weight"`), and the value is an object with the following fields:
+ `outputType`: Specifies the data type to use for the property in the exported data (e.g. "String", "Int", "Float"). For a full-list of supported types and the corresponding type names that can be used in filtering, see {{here}}. If a type is not provided, the export process will determine the type. If a given property is present as multiple types (e.g. one edge has `"weight"` stored as a double, and another edge has it stored as a string), the type will be of Any type. Otherwise, it will be the type of the property as present in edges.
+ `sourcePropertyName`: The name of the property as it exists in the original graph data. If not provided, it is assumed that the key matches the desired `sourcePropertyName`.
+ `multiValueHandling`: Specifies how to handle properties that have multiple values. Can be either "TO\_LIST" to export all values as a list, or "PICK\_FIRST" to export the first value encountered. If not specified, the default value is "PICK\_FIRST".
## Filter syntax
The filter is specified as a JSON object, as follows:
```
{
"vertexFilter": {"string": {
"properties": {"string": {
"outputType": "string",
"sourcePropertyName": "string",
"multiValueHandling": "TO_LIST"|"PICK_FIRST"
}
...}
}
...},
"edgeFilter": {"string": {
"properties": {"string": {
"outputType": "string",
"sourcePropertyName": "string",
"multiValueHandling": "TO_LIST"|"PICK_FIRST"
}
...}
}
...}
}
```
## Filtering label-less vertices and default-labeled edges
Neptune Analytics supports loading vertices without labels and assigns a default label `"edge"` to edges that have no explicit label. The export filter provides special keys to target these entities.
### Filtering label-less vertices
Neptune Analytics supports vertices without labels (vertices loaded with an empty `~label` value or with the `~label` column omitted from the file). To include such vertices in an export, there are three ways:
1. **Omit `vertexFilter` entirely.** If you do not specify a `vertexFilter`, all vertices are exported regardless of label, including label-less ones.
1. **Include `"_NO_LABEL_"` (or equivalently `""`) as a `vertexFilter` key.** Use the sentinel key to target label-less vertices alongside your typed labels:
```
{
"vertexFilter": {
"Person": { "properties": {} },
"_NO_LABEL_": {
"properties": {
"name": { "outputType": "String" }
}
}
}
}
```
Both `"_NO_LABEL_"` and `""` match the same set of vertices — those loaded without a `~label` value.
1. **Use the `"*"` wildcard.** The wildcard key matches all vertex labels including label-less vertices.
**Warning**
If your `vertexFilter` names only explicitly-typed labels (for example, `"Person"` and `"Organization"`) but your graph contains label-less vertices, those vertices are silently excluded from the export. To avoid unexpected data loss, use one of the three approaches listed above.
### Filtering default-labeled edges
Because every edge in Neptune Analytics has a label (edges loaded without an explicit `~label` are assigned the default label `"edge"`), there are three ways to include such edges in an export:
1. **Omit `edgeFilter` entirely.** If you do not specify an `edgeFilter`, all edges are exported regardless of label, including default-labeled ones.
1. **Include `"edge"` as an `edgeFilter` key.** Use the literal string `"edge"` (not `"_NO_LABEL_"`) to target default-labeled edges alongside your typed labels:
```
{
"edgeFilter": {
"route": { "properties": {} },
"edge": { "properties": {} }
}
}
```
1. **Use the `"*"` wildcard.** The wildcard key matches all edge labels including the default `"edge"` label.
**Warning**
If your `edgeFilter` names only explicitly-typed labels (for example, `"route"` and `"knows"`) but your graph contains edges that were loaded without a `~label`, those default-labeled edges are silently excluded from the export. To avoid unexpected data loss, use one of the three approaches listed above.
**Note**
Do **not** use `"_NO_LABEL_"` as an `edgeFilter` key to try to target default-labeled edges. Unlike vertices, edges are never truly label-less — the loader always assigns `"edge"` as the label when `~label` is missing or empty. An `edgeFilter` that names `"_NO_LABEL_"` matches zero edges and exports no edge files.
### Non-existent labels in filters
If a filter key names a label that is not present among the graph's labels, the export fails with a `NotFound` error during initialization, before any data is written. This fail-fast behavior helps catch typos in filter specifications.
**Note**
The sentinel keys `"_NO_LABEL_"`, `""`, `"_VERTEX_ALL_LABELS_"`, `"_EDGE_ALL_LABELS_"`, and `"*"` are always accepted and are not checked against the graph's labels.