From 3e44e328d96c9d9834839974c9339bd4f9adb21c Mon Sep 17 00:00:00 2001 From: Matthew Murray Date: Mon, 3 Jun 2024 12:38:25 -0700 Subject: [PATCH] Add cudf.pandas developer documentation --- .../source/developer_guide/cudf_pandas.md | 96 +++++++++++++++++++ 1 file changed, 96 insertions(+) diff --git a/docs/cudf/source/developer_guide/cudf_pandas.md b/docs/cudf/source/developer_guide/cudf_pandas.md index e2182c1a9dd..31aab569b40 100644 --- a/docs/cudf/source/developer_guide/cudf_pandas.md +++ b/docs/cudf/source/developer_guide/cudf_pandas.md @@ -1 +1,97 @@ # cudf.pandas +The use of the cuDF pandas accelerator mode (`cudf.pandas`) is explained [here](https://docs.rapids.ai/api/cudf/stable/cudf_pandas/). The purpose of this document is to explain how the fast-slow proxy mechanism works and document the internal environment variables used to debug `cudf.pandas`. + +## fast-slow proxy mechanism +`cudf.pandas` works by wrapping each Pandas type (and its corresponding cuDF type) in new proxy types (aka fast-slow proxy types). Because the proxy types wrap both the fast and slow implementations of the original type, we can ensure that computations are first done on the fast version of the proxy type, and if that fails, the slow version of the proxy type. + +### Types +#### Wrapped Types and Proxy Types +The "wrapped" types/classes are the Pandas and cuDF specific types that have been wrapped into proxy types. Wrapped objects and proxy objects are instances of wrapped types and proxy types, respectively. In the snippet below `s1` and `s2` are wrapped objects and `s3` is a fast-slow proxy object. Also note that the module `xpd` is a wrapped module and contains cuDF and Pandas modules as attributes. + ```python + import cudf.pandas + cudf.pandas.install() + import pandas as xpd + + cudf = xpd._fsproxy_fast + pd = xpd._fsproxy_slow + + s1 = cudf.Series([1,2]) + s2 = pd.Series([1,2]) + s3 = xpd.Series([1,2]) + ``` + +#### The Different Kinds of Proxy Types +In `cudf.pandas`, proxy types come in multiple kinds: fast-slow proxy types, callable types, and fast-slow attribute types. + +Fast-slow proxy types come in two flavors: final types and intermediate types. Final types are types for which known operations exist for converting an object of "fast" type to "slow" and vice-versa. For example, `cudf.DataFrame` can be converted to Pandas using the method `to_pandas` and `pd.DataFrame` can be converted to cuDF using the function `cudf.from_pandas`. + +Intermediate types are the types of the results of operations invoked on final types. For example, `DataFrameGroupBy` is a type that will be created during a groupby operation. + +Callable types are ... + +Fast-Slow attribute types ... + +#### Creating New Proxy Types +`_FinalProxy` and `_IntermediateProxy` types are created using the functions `make_final_proxy_type` and `make_intermediate_proxy` type, respectively. Creating a new final type looks like this. + +```python +DataFrame = make_final_proxy_type( + "DataFrame", + cudf.DataFrame, + pd.DataFrame, + fast_to_slow=lambda fast: fast.to_pandas(), + slow_to_fast=cudf.from_pandas, +) +``` + +### The Fallback Mechanism +The `_fast_slow_function_call` is the all-important function that we use to call operations the fast way (using cuDF) and if that fails, the slow way (using Pandas). The is also known as the fallback mechanism. The function looks like this: +```python +def _fast_slow_function_call(func: Callable, *args, **kwargs): + try: + ... + fast_args, fast_kwargs = _fast_arg(args), _fast_arg(kwargs) + result = func(*fast_args, **fast_kwargs) + ... + except Exception: + ... + slow_args, slow_kwargs = _slow_arg(args), _slow_arg(kwargs) + result = func(*slow_args, **slow_kwargs) + ... +``` +As you can see the function attempts to call `func` the fast way using cuDF and if an any `Exception` occurs, it calls the function using Pandas. In essence, this `try-except` is what allows `cudf.pandas` to support 100% of the Pandas API. + +### Using Metaclasses +`cudf.pandas` uses a [metalass](https://docs.python.org/3/glossary.html#term-metaclass) called (`_FastSlowProxyMeta`) to dynamically find class attributes and classmethods of fast-slow proxy types. For example, in the snippet below, the `xpd.Series` type is an instance `_FastSlowProxyMeta`. Therefore we can access the property `_fsproxy_fast` defined in the metaclass. +```python +import cudf.pandas +cudf.pandas.install() +import pandas as xpd + +print(xpd.Series._fsproxy_fast) # output is cudf.core.series.Series +``` + +### Caching + +### Pickling and Unpickling + +## debugging `cudf.pandas` +Several environment variables are available for debugging purposes. + +Setting the environment variable `CUDF_PANDAS_DEBUGGING` produces a warning when the results from cuDF and Pandas differ from one another. For example, the snippet below produces the warning below. +```python +import cudf.pandas +cudf.pandas.install() +import pandas as pd +import numpy as np + +setattr(pd.Series.mean, "_fsproxy_slow", lambda self, *args, **kwargs: np.float64(1)) +s = pd.Series([1,2,3]) +s.mean() +``` +``` +UserWarning: The results from cudf and pandas were different. The exception was +Arrays are not almost equal to 7 decimals + ACTUAL: 1.0 + DESIRED: 2.0. +```