perspective

The Python language bindings for Perspective, a high performance data-visualization and analytics component for the web browser.

A simple example which loads an Apache Arrow and computes a "Group By" operation, returning a new Arrow.

from perspective import Server

client = Server().new_local_client()
table = client.table(arrow_bytes_data)
view = table.view(group_by = ["CounterParty", "Security"])
arrow = view.to_arrow()

Perspective for Python uses the exact same C++ data engine used by the WebAssembly version and Rust version. The library consists of many of the same abstractions and API as in JavaScript, as well as Python-specific data loading support for NumPy, Pandas (and Apache Arrow, as in JavaScript).

Additionally, perspective-python provides a session manager suitable for integration into server systems such as Tornado websockets, AIOHTTP, or Starlette/FastAPI, which allows fully _virtual_ Perspective tables to be interacted with by multiple <perspective-viewer> in a web browser. You can also interact with a Perspective table from python clients, and to that end client libraries are implemented for both Tornado and AIOHTTP.

As <perspective-viewer> will only consume the data necessary to render the current screen, this runtime mode allows _ludicrously-sized_ datasets with instant-load after they've been manifest on the server (at the expense of network latency on UI interaction).

The included PerspectiveWidget allows running such a viewer in JupyterLab in either server or client (via WebAssembly) mode, and the included PerspectiveTornadoHandler makes it simple to extend a Tornado server with virtual Perspective support.

The perspective module exports several tools:

This user's guide provides an overview of the most common ways to use Perspective in Python: the Table API, the JupyterLab widget, and the Tornado handler.

More Examples are available on GitHub.

Installation

perspective-python contains full bindings to the Perspective API, a JupyterLab widget, and a WebSocket handlers for several webserver libraries that allow you to host Perspective using server-side Python.

perspective-python can be installed from PyPI via pip:

pip install perspective-python

Quick Start

A Table can be created from a dataset or a schema, the specifics of which are discussed in the JavaScript section of the user's guide. In Python, however, Perspective supports additional data types that are commonly used when processing data:

  • pandas.DataFrame
  • polars.DataFrame
  • bytes (encoding an Apache Arrow)
  • objects (either extracting a repr or via reference)
  • str (encoding as a CSV)

A Table is created in a similar fashion to its JavaScript equivalent:

from datetime import date, datetime
import numpy as np
import pandas as pd
import perspective

data = pd.DataFrame({
    "int": np.arange(100),
    "float": [i * 1.5 for i in range(100)],
    "bool": [True for i in range(100)],
    "date": [date.today() for i in range(100)],
    "datetime": [datetime.now() for i in range(100)],
    "string": [str(i) for i in range(100)]
})

table = perspective.table(data, index="float")

Likewise, a View can be created via the view() method:

view = table.view(group_by=["float"], filter=[["bool", "==", True]])
column_data = view.to_columns()
row_data = view.to_json()

Pandas and Polars Support

Perspective's Table can be constructed from pandas.DataFrame and polars.DataFrame objects. Internally, this just uses pyarrow::from_pandas, which dictates behavior of this feature including type support.

If the dataframe does not have an index set, an integer-typed column named "index" is created. If you want to preserve the indexing behavior of the dataframe passed into Perspective, simply create the Table with index="index" as a keyword argument. This tells Perspective to once again treat the index as a primary key:

data.set_index("datetime")
table = perspective.table(data, index="index")

Time Zone Handling

When parsing "datetime" strings, times are assumed _local time_ unless an explicit timezone offset is parsed. All "datetime" columns (regardless of input time zone) are _output_ to the user as datetime.datetime objects in _local time_ according to the Python runtime.

This behavior is consistent with Perspective's behavior in JavaScript. For more details, see this in-depth explanation of perspective-python semantics around time zone handling.

Callbacks and Events

perspective.Table allows for on_update and on_delete callbacks to be set—simply call on_update or on_delete with a reference to a function or a lambda without any parameters:

def update_callback():
    print("Updated!")

# set the update callback
on_update_id = view.on_update(update_callback)


def delete_callback():
    print("Deleted!")

# set the delete callback
on_delete_id = view.on_delete(delete_callback)

# set a lambda as a callback
view.on_delete(lambda: print("Deleted x2!"))

If the callback is a named reference to a function, it can be removed with remove_update or remove_delete:

view.remove_update(on_update_id)
view.remove_delete(on_delete_id)

Callbacks defined with a lambda function cannot be removed, as lambda functions have no identifier.

Hosting Table and View instances

Server "hosts" all perspective.Table and perspective.View instances created by its connected Clients. Hosted tables/views can have their methods called from other sources than the Python server, i.e. by a perspective-viewer running in a JavaScript client over the network, interfacing with perspective-python through the websocket API.

The server has full control of all hosted Table and View instances, and can call any public API method on hosted instances. This makes it extremely easy to stream data to a hosted Table using .update():

server = perspective.Server()
client = server.new_local_client()
table = client.table(data, name="data_source")

for i in range(10):
    # updates continue to propagate automatically
    table.update(new_data)

The name provided is important, as it enables Perspective in JavaScript to look up a Table and get a handle to it over the network. Otherwise, name will be assigned randomlu and the Client must look this up with CLient.get_hosted_table_names()

Client/Server Replicated Mode

Using Tornado and PerspectiveTornadoHandler, as well as Perspective's JavaScript library, we can set up "distributed" Perspective instances that allows multiple browser perspective-viewer clients to read from a common perspective-python server, as in the Tornado Example Project.

This architecture works by maintaining two Tables—one on the server, and one on the client that mirrors the server's Table automatically using on_update. All updates to the table on the server are automatically applied to each client, which makes this architecture a natural fit for streaming dashboards and other distributed use-cases. In conjunction with multithreading, distributed Perspective offers consistently high performance over large numbers of clients and large datasets.

_*server.py*_

from perspective import Server
from perspective.hadnlers.tornado import PerspectiveTornadoHandler

# Create an instance of Server, and host a Table
SERVER = Server()
CLIENT = SERVER.new_local_client()

# The Table is exposed at `localhost:8888/websocket` with the name `data_source`
client.table(data, name = "data_source")

app = tornado.web.Application([
    # create a websocket endpoint that the client JavaScript can access
    (r"/websocket", PerspectiveTornadoHandler, {"perspective_server": SERVER})
])

# Start the Tornado server
app.listen(8888)
loop = tornado.ioloop.IOLoop.current()
loop.start()

Instead of calling load(server_table), create a View using server_table and pass that into viewer.load(). This will automatically register an on_update callback that synchronizes state between the server and the client.

_*index.html*_

<perspective-viewer id="viewer" editable></perspective-viewer>

<script type="module">
    // Create a client that expects a Perspective server
    // to accept connections at the specified URL.
    const websocket = await perspective.websocket(
        "ws://localhost:8888/websocket"
    );

    // Get a handle to the Table on the server
    const server_table = await websocket.open_table("data_source_one");

    // Create a new view
    const server_view = await table.view();

    // Create a Table on the client using `perspective.worker()`
    const worker = await perspective.worker();
    const client_table = await worker.table(view);

    // Load the client table in the `<perspective-viewer>`.
    document.getElementById("viewer").load(client_table);
</script>

For a more complex example that offers distributed editing of the server dataset, see client_server_editing.html.

We also provide examples for Starlette/FastAPI and AIOHTTP:

Server-only Mode

The server setup is identical to Distributed Mode above, but instead of creating a view, the client calls load(server_table): In Python, use Server and PerspectiveTornadoHandler to create a websocket server that exposes a Table. In this example, table is a proxy for the Table we created on the server. All API methods are available on _proxies_, the.g.us calling view(), schema(), update() on table will pass those operations to the Python Table, execute the commands, and return the result back to Javascript.

<perspective-viewer id="viewer" editable></perspective-viewer>
const websocket = perspective.websocket("ws://localhost:8888/websocket");
const table = websocket.open_table("data_source");
document.getElementById("viewer").load(table);
  1#  ┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
  2#  ┃ ██████ ██████ ██████       █      █      █      █      █ █▄  ▀███ █       ┃
  3#  ┃ ▄▄▄▄▄█ █▄▄▄▄▄ ▄▄▄▄▄█  ▀▀▀▀▀█▀▀▀▀▀ █ ▀▀▀▀▀█ ████████▌▐███ ███▄  ▀█ █ ▀▀▀▀▀ ┃
  4#  ┃ █▀▀▀▀▀ █▀▀▀▀▀ █▀██▀▀ ▄▄▄▄▄ █ ▄▄▄▄▄█ ▄▄▄▄▄█ ████████▌▐███ █████▄   █ ▄▄▄▄▄ ┃
  5#  ┃ █      ██████ █  ▀█▄       █ ██████      █      ███▌▐███ ███████▄ █       ┃
  6#  ┣━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┫
  7#  ┃ Copyright (c) 2017, the Perspective Authors.                              ┃
  8#  ┃ ╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌╌ ┃
  9#  ┃ This file is part of the Perspective library, distributed under the terms ┃
 10#  ┃ of the [Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0). ┃
 11#  ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛
 12
 13__version__ = "3.7.4"
 14__all__ = [
 15    "_jupyter_labextension_paths",
 16    "Server",
 17    "Client",
 18    "Table",
 19    "View",
 20    "PerspectiveError",
 21    "ProxySession",
 22    "AsyncClient",
 23    "AsyncServer",
 24    "num_cpus",
 25    "set_num_cpus",
 26    "system_info",
 27]
 28
 29__doc__ = """
 30The Python language bindings for [Perspective](https://perspective.finos.org), a
 31high performance data-visualization and analytics component for the web browser.
 32
 33A simple example which loads an [Apache Arrow](https://arrow.apache.org/) and
 34computes a "Group By" operation, returning a new Arrow.
 35
 36```python
 37from perspective import Server
 38
 39client = Server().new_local_client()
 40table = client.table(arrow_bytes_data)
 41view = table.view(group_by = ["CounterParty", "Security"])
 42arrow = view.to_arrow()
 43```
 44
 45Perspective for Python uses the exact same C++ data engine used by the
 46[WebAssembly version](https://docs.rs/perspective-js/latest/perspective_js/) and
 47[Rust version](https://docs.rs/crate/perspective/latest). The library consists
 48of many of the same abstractions and API as in JavaScript, as well as
 49Python-specific data loading support for [NumPy](https://numpy.org/),
 50[Pandas](https://pandas.pydata.org/) (and
 51[Apache Arrow](https://arrow.apache.org/), as in JavaScript).
 52
 53Additionally, `perspective-python` provides a session manager suitable for
 54integration into server systems such as
 55[Tornado websockets](https://www.tornadoweb.org/en/stable/websocket.html),
 56[AIOHTTP](https://docs.aiohttp.org/en/stable/web_quickstart.html#websockets), or
 57[Starlette](https://www.starlette.io/websockets/)/[FastAPI](https://fastapi.tiangolo.com/advanced/websockets/),
 58which allows fully _virtual_ Perspective tables to be interacted with by
 59multiple `<perspective-viewer>` in a web browser. You can also interact with a
 60Perspective table from python clients, and to that end client libraries are
 61implemented for both Tornado and AIOHTTP.
 62
 63As `<perspective-viewer>` will only consume the data necessary to render the
 64current screen, this runtime mode allows _ludicrously-sized_ datasets with
 65instant-load after they've been manifest on the server (at the expense of
 66network latency on UI interaction).
 67
 68The included `PerspectiveWidget` allows running such a viewer in
 69[JupyterLab](https://jupyterlab.readthedocs.io/en/stable/) in either server or
 70client (via WebAssembly) mode, and the included `PerspectiveTornadoHandler`
 71makes it simple to extend a Tornado server with virtual Perspective support.
 72
 73The `perspective` module exports several tools:
 74
 75-   `Server` the constructor for a new isntance of the Perspective data engine.
 76-   The `perspective.widget` module exports `PerspectiveWidget`, the JupyterLab
 77    widget for interactive visualization in a notebook cell.
 78-   The `perspective.handlers` modules exports web frameworks handlers that
 79    interface with a `perspective-client` in JavaScript.
 80    -   `perspective.handlers.tornado.PerspectiveTornadoHandler` for
 81        [Tornado](https://www.tornadoweb.org/)
 82    -   `perspective.handlers.starlette.PerspectiveStarletteHandler` for
 83        [Starlette](https://www.starlette.io/) and
 84        [FastAPI](https://fastapi.tiangolo.com)
 85    -   `perspective.handlers.aiohttp.PerspectiveAIOHTTPHandler` for
 86        [AIOHTTP](https://docs.aiohttp.org),
 87
 88This user's guide provides an overview of the most common ways to use
 89Perspective in Python: the `Table` API, the JupyterLab widget, and the Tornado
 90handler.
 91
 92[More Examples](https://github.com/finos/perspective/tree/master/examples) are
 93available on GitHub.
 94
 95## Installation
 96
 97`perspective-python` contains full bindings to the Perspective API, a JupyterLab
 98widget, and a WebSocket handlers for several webserver libraries that allow you
 99to host Perspective using server-side Python.
100
101`perspective-python` can be installed from [PyPI](https://pypi.org) via `pip`:
102
103```bash
104pip install perspective-python
105```
106
107## Quick Start
108
109A `Table` can be created from a dataset or a schema, the specifics of which are
110[discussed](#loading-data-with-table) in the JavaScript section of the user's
111guide. In Python, however, Perspective supports additional data types that are
112commonly used when processing data:
113
114-   `pandas.DataFrame`
115-   `polars.DataFrame`
116-   `bytes` (encoding an Apache Arrow)
117-   `objects` (either extracting a repr or via reference)
118-   `str` (encoding as a CSV)
119
120A `Table` is created in a similar fashion to its JavaScript equivalent:
121
122```python
123from datetime import date, datetime
124import numpy as np
125import pandas as pd
126import perspective
127
128data = pd.DataFrame({
129    "int": np.arange(100),
130    "float": [i * 1.5 for i in range(100)],
131    "bool": [True for i in range(100)],
132    "date": [date.today() for i in range(100)],
133    "datetime": [datetime.now() for i in range(100)],
134    "string": [str(i) for i in range(100)]
135})
136
137table = perspective.table(data, index="float")
138```
139
140Likewise, a `View` can be created via the `view()` method:
141
142```python
143view = table.view(group_by=["float"], filter=[["bool", "==", True]])
144column_data = view.to_columns()
145row_data = view.to_json()
146```
147
148#### Pandas and Polars Support
149
150Perspective's `Table` can be constructed from `pandas.DataFrame` and
151`polars.DataFrame` objects. Internally, this just uses
152[`pyarrow::from_pandas`](https://arrow.apache.org/docs/python/pandas.html),
153which dictates behavior of this feature including type support.
154
155If the dataframe does not have an index set, an integer-typed column named
156`"index"` is created. If you want to preserve the indexing behavior of the
157dataframe passed into Perspective, simply create the `Table` with
158`index="index"` as a keyword argument. This tells Perspective to once again
159treat the index as a primary key:
160
161```python
162data.set_index("datetime")
163table = perspective.table(data, index="index")
164```
165
166#### Time Zone Handling
167
168When parsing `"datetime"` strings, times are assumed _local time_ unless an
169explicit timezone offset is parsed. All `"datetime"` columns (regardless of
170input time zone) are _output_ to the user as `datetime.datetime` objects in
171_local time_ according to the Python runtime.
172
173This behavior is consistent with Perspective's behavior in JavaScript. For more
174details, see this in-depth
175[explanation](https://github.com/finos/perspective/pull/867) of
176`perspective-python` semantics around time zone handling.
177
178#### Callbacks and Events
179
180`perspective.Table` allows for `on_update` and `on_delete` callbacks to be
181set—simply call `on_update` or `on_delete` with a reference to a function or a
182lambda without any parameters:
183
184```python
185def update_callback():
186    print("Updated!")
187
188# set the update callback
189on_update_id = view.on_update(update_callback)
190
191
192def delete_callback():
193    print("Deleted!")
194
195# set the delete callback
196on_delete_id = view.on_delete(delete_callback)
197
198# set a lambda as a callback
199view.on_delete(lambda: print("Deleted x2!"))
200```
201
202If the callback is a named reference to a function, it can be removed with
203`remove_update` or `remove_delete`:
204
205```python
206view.remove_update(on_update_id)
207view.remove_delete(on_delete_id)
208```
209
210Callbacks defined with a lambda function cannot be removed, as lambda functions
211have no identifier.
212
213### Hosting `Table` and `View` instances
214
215`Server` "hosts" all `perspective.Table` and `perspective.View` instances
216created by its connected `Client`s. Hosted tables/views can have their methods
217called from other sources than the Python server, i.e. by a `perspective-viewer`
218running in a JavaScript client over the network, interfacing with
219`perspective-python` through the websocket API.
220
221The server has full control of all hosted `Table` and `View` instances, and can
222call any public API method on hosted instances. This makes it extremely easy to
223stream data to a hosted `Table` using `.update()`:
224
225```python
226server = perspective.Server()
227client = server.new_local_client()
228table = client.table(data, name="data_source")
229
230for i in range(10):
231    # updates continue to propagate automatically
232    table.update(new_data)
233```
234
235The `name` provided is important, as it enables Perspective in JavaScript to
236look up a `Table` and get a handle to it over the network. Otherwise, `name`
237will be assigned randomlu and the `Client` must look this up with
238`CLient.get_hosted_table_names()`
239
240### Client/Server Replicated Mode
241
242Using Tornado and
243[`PerspectiveTornadoHandler`](python.md#perspectivetornadohandler), as well as
244`Perspective`'s JavaScript library, we can set up "distributed" Perspective
245instances that allows multiple browser `perspective-viewer` clients to read from
246a common `perspective-python` server, as in the
247[Tornado Example Project](https://github.com/finos/perspective/tree/master/examples/python-tornado).
248
249This architecture works by maintaining two `Tables`—one on the server, and one
250on the client that mirrors the server's `Table` automatically using `on_update`.
251All updates to the table on the server are automatically applied to each client,
252which makes this architecture a natural fit for streaming dashboards and other
253distributed use-cases. In conjunction with [multithreading](#multi-threading),
254distributed Perspective offers consistently high performance over large numbers
255of clients and large datasets.
256
257_*server.py*_
258
259```python
260from perspective import Server
261from perspective.hadnlers.tornado import PerspectiveTornadoHandler
262
263# Create an instance of Server, and host a Table
264SERVER = Server()
265CLIENT = SERVER.new_local_client()
266
267# The Table is exposed at `localhost:8888/websocket` with the name `data_source`
268client.table(data, name = "data_source")
269
270app = tornado.web.Application([
271    # create a websocket endpoint that the client JavaScript can access
272    (r"/websocket", PerspectiveTornadoHandler, {"perspective_server": SERVER})
273])
274
275# Start the Tornado server
276app.listen(8888)
277loop = tornado.ioloop.IOLoop.current()
278loop.start()
279```
280
281Instead of calling `load(server_table)`, create a `View` using `server_table`
282and pass that into `viewer.load()`. This will automatically register an
283`on_update` callback that synchronizes state between the server and the client.
284
285_*index.html*_
286
287```html
288<perspective-viewer id="viewer" editable></perspective-viewer>
289
290<script type="module">
291    // Create a client that expects a Perspective server
292    // to accept connections at the specified URL.
293    const websocket = await perspective.websocket(
294        "ws://localhost:8888/websocket"
295    );
296
297    // Get a handle to the Table on the server
298    const server_table = await websocket.open_table("data_source_one");
299
300    // Create a new view
301    const server_view = await table.view();
302
303    // Create a Table on the client using `perspective.worker()`
304    const worker = await perspective.worker();
305    const client_table = await worker.table(view);
306
307    // Load the client table in the `<perspective-viewer>`.
308    document.getElementById("viewer").load(client_table);
309</script>
310```
311
312For a more complex example that offers distributed editing of the server
313dataset, see
314[client_server_editing.html](https://github.com/finos/perspective/blob/master/examples/python-tornado/client_server_editing.html).
315
316We also provide examples for Starlette/FastAPI and AIOHTTP:
317
318-   [Starlette Example Project](https://github.com/finos/perspective/tree/master/examples/python-starlette).
319-   [AIOHTTP Example Project](https://github.com/finos/perspective/tree/master/examples/python-aiohttp).
320
321### Server-only Mode
322
323The server setup is identical to [Distributed Mode](#distributed-mode) above,
324but instead of creating a view, the client calls `load(server_table)`: In
325Python, use `Server` and `PerspectiveTornadoHandler` to create a websocket
326server that exposes a `Table`. In this example, `table` is a proxy for the
327`Table` we created on the server. All API methods are available on _proxies_,
328the.g.us calling `view()`, `schema()`, `update()` on `table` will pass those
329operations to the Python `Table`, execute the commands, and return the result
330back to Javascript.
331
332```html
333<perspective-viewer id="viewer" editable></perspective-viewer>
334```
335
336```javascript
337const websocket = perspective.websocket("ws://localhost:8888/websocket");
338const table = websocket.open_table("data_source");
339document.getElementById("viewer").load(table);
340```
341
342"""
343
344
345import functools
346
347from .perspective import (
348    Client,
349    PerspectiveError,
350    ProxySession,
351    Server,
352    AsyncServer,
353    AsyncClient,
354    # NOTE: these are classes without constructors,
355    # so we import them just for type hinting
356    Table,  # noqa: F401
357    View,  # noqa: F401
358    num_cpus,
359    set_num_cpus,
360)
361
362
363GLOBAL_SERVER = Server()
364GLOBAL_CLIENT = GLOBAL_SERVER.new_local_client()
365
366
367@functools.wraps(Client.table)
368def table(*args, **kwargs):
369    return GLOBAL_CLIENT.table(*args, **kwargs)
370
371
372@functools.wraps(Client.open_table)
373def open_table(*args, **kwargs):
374    return GLOBAL_CLIENT.table(*args, **kwargs)
375
376
377@functools.wraps(Client.get_hosted_table_names)
378def get_hosted_table_names(*args, **kwargs):
379    return GLOBAL_CLIENT.get_hosted_table_names(*args, **kwargs)
380
381
382@functools.wraps(Client.system_info)
383def system_info(*args, **kwargs):
384    return GLOBAL_CLIENT.system_info(*args, **kwargs)
385
386
387def _jupyter_labextension_paths():
388    """
389    Read by `jupyter labextension develop`
390    @private
391    """
392    return [{"src": "labextension", "dest": "@finos/perspective-jupyterlab"}]
class Server:

An instance of a Perspective server. Each Server instance is separate, and does not share Table (or other) data with other Servers.

Arguments

  • on_poll_request A callback function which the Server will invoke when there are updates that need to be flushed, after which you must _eventually_ call Server.poll (or else no updates will be processed). This optimization allows batching updates, depending on context.
def new_local_client(self, /):

Create a new Client instance bound to this Server directly.

def new_session(self, /, response_cb):

Create a new Session bound to this Server.

Server.new_session only needs to be called if you've implemented a custom Perspective ['Client]/[Server`] transport.

def poll(self, /):

Flush pending updates to this Server, including notifications to View.on_update callbacks.

Server.poll only needs to be called if you've implemented a custom Perspective Server and provided the on_poll_request constructor keyword argument.

Calling Session.poll may result in the send_response parameter which was used to construct this (or other) Session to fire. Whenever a Session.handle_request method is invoked for a perspective_server::Server, at least one Session.poll should be scheduled to clear other clients message queues.

poll() _must_ be called after Table.update or Table.remove and on_poll_request is notified, or the changes will not be applied.

class Client:

An instance of a Client is a connection to a single Server, whether locally in-memory or remote over some transport like a WebSocket.

Client and Perspective objects derived from it have _synchronous_ APIs, suitable for use in a repl or script context where this is the _only_ Client connected to its Server. If you want to integrate with a Web framework or otherwise connect multiple clients, use AsyncClient.

def from_server(server):

Create a new Client instance bound to a specific in-process Server (e.g. generally _not_ the global Server).

def handle_response(self, /, response):

Handle a message from the external message queue. Client.handle_response is part of the low-level message-handling API necessary to implement new transports for a Client connection to a local-or-remote Server, and doesn't generally need to be called directly by "users" of a Client once connected.

def table(self, /, input, limit=None, index=None, name=None, format=None):

Creates a new Table from either a _schema_ or _data_.

The Client.table factory function can be initialized with either a _schema_ (see Table.schema), or data in one of these formats:

  • Apache Arrow
  • CSV
  • JSON row-oriented
  • JSON column-oriented
  • NDJSON

When instantiated with _data_, the schema is inferred from this data. While this is convenient, inferrence is sometimes imperfect e.g. when the input is empty, null or ambiguous. For these cases, Client.table can first be instantiated with a explicit schema.

When instantiated with a _schema_, the resulting Table is empty but with known column names and column types. When subsqeuently populated with Table.update, these columns will be _coerced_ to the schema's type. This behavior can be useful when Client.table's column type inferences doesn't work.

The resulting Table is _virtual_, and invoking its methods dispatches events to the perspective_server::Server this Client connects to, where the data is stored and all calculation occurs.

Arguments

  • arg - Either _schema_ or initialization _data_.
  • options - Optional configuration which provides one of:
    • limit - The max number of rows the resulting Table can store.
    • index - The column name to use as an _index_ column. If this Table is being instantiated by _data_, this column name must be present in the data.
    • name - The name of the table. This will be generated if it is not provided.
    • format - The explicit format of the input data, can be one of "json", "columns", "csv" or "arrow". This overrides language-specific type dispatch behavior, which allows stringified and byte array alternative inputs.

Python Examples

Load a CSV from a str:

table = client.table("x,y\n1,2\n3,4")
def open_table(self, /, name):

Opens a Table that is hosted on the perspective_server::Server that is connected to this Client.

The name property of TableInitOptions is used to identify each Table. Table names can be looked up for each Client via Client.get_hosted_table_names.

Python Examples

table =  client.open_table("table_one");
def get_hosted_table_names(self, /):

Retrieves the names of all tables that this client has access to.

name is a string identifier unique to the Table (per Client), which can be used in conjunction with Client.open_table to get a Table instance without the use of Client.table constructor directly (e.g., one created by another Client).

Python Examples

tables = client.get_hosted_table_names();
def on_hosted_tables_update(self, /, callback):

Register a callback which is invoked whenever Client.table (on this Client) or Table.delete (on a Table belinging to this Client) are called.

def remove_hosted_tables_update(self, /, callback_id):

Remove a callback previously registered via Client.on_hosted_tables_update.

def system_info(self, /):

Provides the SystemInfo struct, implementation-specific metadata about the perspective_server.Server runtime such as Memory and CPU usage.

def terminate(self, /):

Terminates this Client, cleaning up any View handles the Client has open as well as its callbacks.

class Table:

Table is Perspective's columnar data frame, analogous to a Pandas/Polars DataFrame or Apache Arrow, supporting append & in-place updates, removal by index, and update notifications.

A Table contains columns, each of which have a unique name, are strongly and consistently typed, and contains rows of data conforming to the column's type. Each column in a Table must have the same number of rows, though not every row must contain data; null-values are used to indicate missing values in the dataset. The schema of a Table is _immutable after creation_, which means the column names and data types cannot be changed after the Table has been created. Columns cannot be added or deleted after creation either, but a View can be used to select an arbitrary set of columns from the Table.

def get_index(self, /):

Returns the name of the index column for the table.

Python Examples

table = perspective.table("x,y\n1,2\n3,4", index="x");
index = client.get_index()
def get_client(self, /):

Get a copy of the Client this Table came from.

def get_limit(self, /):

Returns the user-specified row limit for this table.

def get_name(self, /):

Returns the user-specified name for this table, or the auto-generated name if a name was not specified when the table was created.

def clear(self, /):

Removes all the rows in the Table, but preserves everything else including the schema, index, and any callbacks or registered View instances.

Calling Table.clear, like Table.update and Table.remove, will trigger an update event to any registered listeners via View.on_update.

def columns(self, /):

Returns the column names of this Table in "natural" order (the ordering implied by the input format).

# Python Examples

columns = table.columns()
def delete(self, /, lazy=False):

Delete this Table and cleans up associated resources.

Tables do not stop consuming resources or processing updates when they are garbage collected in their host language - you must call this method to reclaim these.

Arguments

  • options An options dictionary.
    • lazy Whether to delete this Table _lazily_. When false (the default), the delete will occur immediately, assuming it has no View instances registered to it (which must be deleted first, otherwise this method will throw an error). When true, the Table will only be marked for deltion once its View dependency count reaches 0.

Python Examples

table = client.table("x,y\n1,2\n3,4")

# ...

table.delete(lazy=True)
def make_port(self, /):

Create a unique channel ID on this Table, which allows View::on_update callback calls to be associated with the Table::update which caused them.

def on_delete(self, /, callback):

Register a callback which is called exactly once, when this Table is deleted with the Table.delete method.

Table.on_delete resolves when the subscription message is sent, not when the _delete_ event occurs.

def remove(self, /, input, format=None):
def remove_delete(self, /, callback_id):

Removes a listener with a given ID, as returned by a previous call to Table.on_delete.

def schema(self, /):

Returns a table's Schema, a mapping of column names to column types.

The mapping of a Table's column names to data types is referred to as a Schema. Each column has a unique name and a data type, one of:

  • "boolean" - A boolean type
  • "date" - A timesonze-agnostic date type (month/day/year)
  • "datetime" - A millisecond-precision datetime type in the UTC timezone
  • "float" - A 64 bit float
  • "integer" - A signed 32 bit integer (the integer type supported by JavaScript)
  • "string" - A String data type (encoded internally as a _dictionary_)

Note that all Table columns are _nullable_, regardless of the data type.

def validate_expressions(self, /, expression):

Validates the given expressions.

def view(self, /, **config):

Create a new View from this table with a specified ViewConfigUpdate.

See View struct.

Examples

view view = table.view(
    columns=["Sales"],
    aggregates={"Sales": "sum"},
    group_by=["Region", "State"],
)
def size(self, /):

Returns the number of rows in a Table.

def replace(self, /, input, format=None):

Removes all the rows in the Table, but preserves everything else including the schema, index, and any callbacks or registered View instances.

Calling Table.clear, like Table.update and Table.remove, will trigger an update event to any registered listeners via View.on_update.

def update(self, /, input, port_id=None, format=None):

Updates the rows of this table and any derived View instances.

Calling Table.update will trigger the View.on_update callbacks register to derived View, and the call itself will not resolve until _all_ derived View's are notified.

When updating a Table with an index, Table.update supports partial updates, by omitting columns from the update data.

Arguments

  • input - The input data for this Table. The schema of a Table is immutable after creation, so this method cannot be called with a schema.
  • options - Options for this update step - see perspective_client.UpdateOptions. ```
class View:

The View struct is Perspective's query and serialization interface. It represents a query on the Table's dataset and is always created from an existing Table instance via the Table.view method.

Views are immutable with respect to the arguments provided to the Table.view method; to change these parameters, you must create a new View on the same Table. However, each View is _live_ with respect to the Table's data, and will (within a conflation window) update with the latest state as its parent Table updates, including incrementally recalculating all aggregates, pivots, filters, etc. View query parameters are composable, in that each parameter works independently _and_ in conjunction with each other, and there is no limit to the number of pivots, filters, etc. which can be applied.

To construct a View, call the Table.view factory method. A Table can have as many Views associated with it as you need - Perspective conserves memory by relying on a single Table to power multiple Views concurrently.

def column_paths(self, /):

Returns an array of strings containing the column paths of the View without any of the source columns.

A column path shows the columns that a given cell belongs to after pivots are applied.

def to_columns_string(self, /, **window):

Renders this View as a column-oriented JSON string. Useful if you want to save additional round trip serialize/deserialize cycles.

def to_json_string(self, /, **window):

Renders this View as a row-oriented JSON string.

def to_ndjson(self, /, **window):

Renders this View as an NDJSON formatted String.

def to_records(self, /, **window):

Renders this View as a row-oriented Python list.

def to_json(self, /, **window):

Renders this View as a row-oriented Python list.

def to_columns(self, /, **window):

Renders this View as a column-oriented Python dict.

def to_csv(self, /, **window):

Renders this View as a CSV String in a standard format.

def to_dataframe(self, /, **window):

Renders this View as a pandas.DataFrame.

def to_pandas(self, /, **window):

Renders this View as a pandas.DataFrame.

def to_polars(self, /, **window):

Renders this View as a polars.DataFrame.

def to_arrow(self, /, **window):

Renders this View as the Apache Arrow data format.

Arguments

def delete(self, /):

Delete this View and clean up all resources associated with it. View objects do not stop consuming resources or processing updates when they are garbage collected - you must call this method to reclaim these.

def expand(self, /, index):
def collapse(self, /, index):
def dimensions(self, /):

Returns this View's _dimensions_, row and column count, as well as those of the crate.Table from which it was derived.

  • num_table_rows - The number of rows in the underlying crate.Table.
  • num_table_columns - The number of columns in the underlying crate.Table (including the index column if this crate.Table was constructed with one).
  • num_view_rows - The number of rows in this View. If this View has a group_by clause, num_view_rows will also include aggregated rows.
  • num_view_columns - The number of columns in this View. If this View has a split_by clause, num_view_columns will include all _column paths_, e.g. the number of columns clause times the number of split_by groups.
def expression_schema(self, /):

The expression schema of this View, which contains only the expressions created on this View. See View.schema for details.

def get_config(self, /):

A copy of the ViewConfig object passed to the Table.view method which created this View.

def get_min_max(self, /, column_name):

Calculates the [min, max] of the leaf nodes of a column column_name.

Returns

A tuple of [min, max], whose types are column and aggregate dependent.

def num_rows(self, /):

The number of aggregated rows in this View. This is affected by the "group_by" configuration parameter supplied to this view's contructor.

Returns

The number of aggregated rows.

def schema(self, /):

The schema of this View.

The View schema differs from the schema returned by Table.schema; it may have different column names due to expressions or columns configs, or it maye have _different column types_ due to the application og group_by and aggregates config. You can think of Table.schema as the _input_ schema and View.schema as the _output_ schema of a Perspective pipeline.

def on_delete(self, /, callback):

Register a callback with this View. Whenever the View is deleted, this callback will be invoked.

def remove_delete(self, /, callback_id):

Unregister a previously registered View.on_delete callback.

def on_update(self, /, callback, mode=None):

Register a callback with this View. Whenever the view's underlying table emits an update, this callback will be invoked with an object containing port_id, indicating which port the update fired on, and optionally delta, which is the new data that was updated for each cell or each row.

Arguments

  • on_update - A callback function invoked on update, which receives an object with two keys: port_id, indicating which port the update was triggered on, and delta, whose value is dependent on the mode parameter.
  • options - If this is provided as OnUpdateOptions { mode: Some(OnUpdateMode::Row) }, then delta is an Arrow of the updated rows. Otherwise delta will be Option.None.
def remove_update(self, /, callback_id):

Unregister a previously registered update callback with this View.

Arguments

  • id - A callback id as returned by a recipricol call to View.on_update.

Examples

let callback = |_| async { print!("Updated!") };
let cid = view.on_update(callback, OnUpdateOptions::default()).await?;
view.remove_update(cid).await?;
PerspectiveError = <class 'perspective.PyPerspectiveError'>
class ProxySession:
def handle_request(self, /, data):
def handle_request_async(self, /, data):
def close(self, /):
def num_cpus():

Returns the number of threads the internal threadpool will use.

def set_num_cpus(num_cpus):

Set the number of threads the internal threadpool will use. Can also be set with NUM_OMP_THREADS environment variable.

def system_info(self, /):

Provides the SystemInfo struct, implementation-specific metadata about the perspective_server.Server runtime such as Memory and CPU usage.