10 Minutes to cuDF and Dask cuDF#
Modelled after 10 Minutes to Pandas, this is a short introduction to cuDF and Dask cuDF, geared mainly towards new users.
What are these Libraries?#
cuDF is a Python GPU DataFrame library (built on the Apache Arrow columnar memory format) for loading, joining, aggregating, filtering, and otherwise manipulating tabular data using a DataFrame style API in the style of pandas.
Dask is a flexible library for parallel computing in Python that makes scaling out your workflow smooth and simple. On the CPU, Dask uses Pandas to execute operations in parallel on DataFrame partitions.
Dask cuDF extends Dask where necessary to allow its DataFrame partitions to be processed using cuDF GPU DataFrames instead of Pandas DataFrames. For instance, when you call dask_cudf.read_csv(...), your cluster’s GPUs do the work of parsing the CSV file(s) by calling cudf.read_csv().
When to use cuDF and Dask cuDF#
If your workflow is fast enough on a single GPU or your data comfortably fits in memory on a single GPU, you would want to use cuDF. If you want to distribute your workflow across multiple GPUs, have more data than you can fit in memory on a single GPU, or want to analyze data spread across many files at once, you would want to use Dask cuDF.
import os
import cupy as cp
import dask_cudf
import pandas as pd
import cudf
cp.random.seed(12)
#### Portions of this were borrowed and adapted from the
#### cuDF cheatsheet, existing cuDF documentation,
#### and 10 Minutes to Pandas.
Object Creation#
Creating a cudf.Series and dask_cudf.Series.
s = cudf.Series([1, 2, 3, None, 4])
s
0 1
1 2
2 3
3 <NA>
4 4
dtype: int64
ds = dask_cudf.from_cudf(s, npartitions=2)
# Note the call to head here to show the first few entries, unlike
# cuDF objects, Dask-cuDF objects do not have a printing
# representation that shows values since they may not be in local
# memory.
ds.head(n=3)
0 1
1 2
2 3
dtype: int64
Creating a cudf.DataFrame and a dask_cudf.DataFrame by specifying values for each column.
df = cudf.DataFrame(
{
"a": list(range(20)),
"b": list(reversed(range(20))),
"c": list(range(20)),
}
)
df
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 1 | 1 | 18 | 1 |
| 2 | 2 | 17 | 2 |
| 3 | 3 | 16 | 3 |
| 4 | 4 | 15 | 4 |
| 5 | 5 | 14 | 5 |
| 6 | 6 | 13 | 6 |
| 7 | 7 | 12 | 7 |
| 8 | 8 | 11 | 8 |
| 9 | 9 | 10 | 9 |
| 10 | 10 | 9 | 10 |
| 11 | 11 | 8 | 11 |
| 12 | 12 | 7 | 12 |
| 13 | 13 | 6 | 13 |
| 14 | 14 | 5 | 14 |
| 15 | 15 | 4 | 15 |
| 16 | 16 | 3 | 16 |
| 17 | 17 | 2 | 17 |
| 18 | 18 | 1 | 18 |
| 19 | 19 | 0 | 19 |
Now we will convert our cuDF dataframe into a Dask-cuDF equivalent. Here we call out a key difference: to inspect the data we must call a method (here .head() to look at the first few values). In the general case (see the end of this notebook), the data in ddf will be distributed across multiple GPUs.
In this small case, we could call ddf.compute() to obtain a cuDF object from the Dask-cuDF object. In general, we should avoid calling .compute() on large dataframes, and restrict ourselves to using it when we have some (relatively) small postprocessed result that we wish to inspect. Hence, throughout this notebook we will generally call .head() to inspect the first few values of a Dask-cuDF dataframe, occasionally calling out places where we use .compute() and why.
To understand more of the differences between how cuDF and Dask cuDF behave here, visit the 10 Minutes to Dask tutorial after this one.
ddf = dask_cudf.from_cudf(df, npartitions=2)
ddf.head()
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 1 | 1 | 18 | 1 |
| 2 | 2 | 17 | 2 |
| 3 | 3 | 16 | 3 |
| 4 | 4 | 15 | 4 |
Creating a cudf.DataFrame from a pandas Dataframe and a dask_cudf.Dataframe from a cudf.Dataframe.
Note that best practice for using dask-cuDF is to read data directly into a dask_cudf.DataFrame with read_csv or other builtin I/O routines (discussed below).
pdf = pd.DataFrame({"a": [0, 1, 2, 3], "b": [0.1, 0.2, None, 0.3]})
gdf = cudf.DataFrame.from_pandas(pdf)
gdf
| a | b | |
|---|---|---|
| 0 | 0 | 0.1 |
| 1 | 1 | 0.2 |
| 2 | 2 | <NA> |
| 3 | 3 | 0.3 |
dask_gdf = dask_cudf.from_cudf(gdf, npartitions=2)
dask_gdf.head(n=2)
| a | b | |
|---|---|---|
| 0 | 0 | 0.1 |
| 1 | 1 | 0.2 |
Viewing Data#
Viewing the top rows of a GPU dataframe.
df.head(2)
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 1 | 1 | 18 | 1 |
ddf.head(2)
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 1 | 1 | 18 | 1 |
Sorting by values.
df.sort_values(by="b")
| a | b | c | |
|---|---|---|---|
| 19 | 19 | 0 | 19 |
| 18 | 18 | 1 | 18 |
| 17 | 17 | 2 | 17 |
| 16 | 16 | 3 | 16 |
| 15 | 15 | 4 | 15 |
| 14 | 14 | 5 | 14 |
| 13 | 13 | 6 | 13 |
| 12 | 12 | 7 | 12 |
| 11 | 11 | 8 | 11 |
| 10 | 10 | 9 | 10 |
| 9 | 9 | 10 | 9 |
| 8 | 8 | 11 | 8 |
| 7 | 7 | 12 | 7 |
| 6 | 6 | 13 | 6 |
| 5 | 5 | 14 | 5 |
| 4 | 4 | 15 | 4 |
| 3 | 3 | 16 | 3 |
| 2 | 2 | 17 | 2 |
| 1 | 1 | 18 | 1 |
| 0 | 0 | 19 | 0 |
ddf.sort_values(by="b").head()
| a | b | c | |
|---|---|---|---|
| 19 | 19 | 0 | 19 |
| 18 | 18 | 1 | 18 |
| 17 | 17 | 2 | 17 |
| 16 | 16 | 3 | 16 |
| 15 | 15 | 4 | 15 |
Selecting a Column#
Selecting a single column, which initially yields a cudf.Series or dask_cudf.Series. Calling compute results in a cudf.Series (equivalent to df.a).
df["a"]
0 0
1 1
2 2
3 3
4 4
5 5
6 6
7 7
8 8
9 9
10 10
11 11
12 12
13 13
14 14
15 15
16 16
17 17
18 18
19 19
Name: a, dtype: int64
ddf["a"].head()
0 0
1 1
2 2
3 3
4 4
Name: a, dtype: int64
Selecting Rows by Label#
Selecting rows from index 2 to index 5 from columns ‘a’ and ‘b’.
df.loc[2:5, ["a", "b"]]
| a | b | |
|---|---|---|
| 2 | 2 | 17 |
| 3 | 3 | 16 |
| 4 | 4 | 15 |
| 5 | 5 | 14 |
ddf.loc[2:5, ["a", "b"]].head()
| a | b | |
|---|---|---|
| 2 | 2 | 17 |
| 3 | 3 | 16 |
| 4 | 4 | 15 |
| 5 | 5 | 14 |
Selecting Rows by Position#
Selecting via integers and integer slices, like numpy/pandas. Note that this functionality is not available for Dask-cuDF DataFrames.
df.iloc[0]
a 0
b 19
c 0
Name: 0, dtype: int64
df.iloc[0:3, 0:2]
| a | b | |
|---|---|---|
| 0 | 0 | 19 |
| 1 | 1 | 18 |
| 2 | 2 | 17 |
You can also select elements of a DataFrame or Series with direct index access.
df[3:5]
| a | b | c | |
|---|---|---|---|
| 3 | 3 | 16 | 3 |
| 4 | 4 | 15 | 4 |
s[3:5]
3 <NA>
4 4
dtype: int64
Boolean Indexing#
Selecting rows in a DataFrame or Series by direct Boolean indexing.
df[df.b > 15]
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 1 | 1 | 18 | 1 |
| 2 | 2 | 17 | 2 |
| 3 | 3 | 16 | 3 |
ddf[ddf.b > 15].head(n=3)
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 1 | 1 | 18 | 1 |
| 2 | 2 | 17 | 2 |
Selecting values from a DataFrame where a Boolean condition is met, via the query API.
df.query("b == 3")
| a | b | c | |
|---|---|---|---|
| 16 | 16 | 3 | 16 |
Note here we call compute() rather than head() on the Dask-cuDF dataframe since we are happy that the number of matching rows will be small (and hence it is reasonable to bring the entire result back).
ddf.query("b == 3").compute()
| a | b | c | |
|---|---|---|---|
| 16 | 16 | 3 | 16 |
You can also pass local variables to Dask-cuDF queries, via the local_dict keyword. With standard cuDF, you may either use the local_dict keyword or directly pass the variable via the @ keyword. Supported logical operators include >, <, >=, <=, ==, and !=.
cudf_comparator = 3
df.query("b == @cudf_comparator")
| a | b | c | |
|---|---|---|---|
| 16 | 16 | 3 | 16 |
dask_cudf_comparator = 3
ddf.query("b == @val", local_dict={"val": dask_cudf_comparator}).compute()
| a | b | c | |
|---|---|---|---|
| 16 | 16 | 3 | 16 |
Using the isin method for filtering.
df[df.a.isin([0, 5])]
| a | b | c | |
|---|---|---|---|
| 0 | 0 | 19 | 0 |
| 5 | 5 | 14 | 5 |
MultiIndex#
cuDF supports hierarchical indexing of DataFrames using MultiIndex. Grouping hierarchically (see Grouping below) automatically produces a DataFrame with a MultiIndex.
arrays = [["a", "a", "b", "b"], [1, 2, 3, 4]]
tuples = list(zip(*arrays))
idx = cudf.MultiIndex.from_tuples(tuples)
idx
MultiIndex([('a', 1),
('a', 2),
('b', 3),
('b', 4)],
)
This index can back either axis of a DataFrame.
gdf1 = cudf.DataFrame(
{"first": cp.random.rand(4), "second": cp.random.rand(4)}
)
gdf1.index = idx
gdf1
| first | second | ||
|---|---|---|---|
| a | 1 | 0.082654 | 0.967955 |
| 2 | 0.399417 | 0.441425 | |
| b | 3 | 0.784297 | 0.793582 |
| 4 | 0.070303 | 0.271711 |
gdf2 = cudf.DataFrame(
{"first": cp.random.rand(4), "second": cp.random.rand(4)}
).T
gdf2.columns = idx
gdf2
| a | b | |||
|---|---|---|---|---|
| 1 | 2 | 3 | 4 | |
| first | 0.343382 | 0.003700 | 0.20043 | 0.581614 |
| second | 0.907812 | 0.101512 | 0.24179 | 0.224180 |
Accessing values of a DataFrame with a MultiIndex, both with .loc
gdf1.loc[("b", 3)]
first 0.784297
second 0.793582
Name: ('b', 3), dtype: float64
And .iloc
gdf1.iloc[0:2]
| first | second | ||
|---|---|---|---|
| a | 1 | 0.082654 | 0.967955 |
| 2 | 0.399417 | 0.441425 |
Missing Data#
Missing data can be replaced by using the fillna method.
s.fillna(999)
0 1
1 2
2 3
3 999
4 4
dtype: int64
ds.fillna(999).head(n=3)
0 1
1 2
2 3
dtype: int64
Stats#
Calculating descriptive statistics for a Series.
s.mean(), s.var()
(np.float64(2.5), np.float64(1.6666666666666667))
This serves as a prototypical example of when we might want to call .compute(). The result of computing the mean and variance is a single number in each case, so it is definitely reasonable to look at the entire result!
ds.mean().compute(), ds.var().compute()
(np.float64(2.5), np.float64(1.6666666666666667))
Applymap#
Applying functions to a Series. Note that applying user defined functions directly with Dask cuDF is not yet implemented. For now, you can use map_partitions to apply a function to each partition of the distributed dataframe.
def add_ten(num):
return num + 10
df["a"].apply(add_ten)
0 10
1 11
2 12
3 13
4 14
5 15
6 16
7 17
8 18
9 19
10 20
11 21
12 22
13 23
14 24
15 25
16 26
17 27
18 28
19 29
Name: a, dtype: int64
ddf["a"].map_partitions(add_ten).head()
0 10
1 11
2 12
3 13
4 14
Name: a, dtype: int64
Histogramming#
Counting the number of occurrences of each unique value of variable.
df.a.value_counts()
a
15 1
10 1
18 1
2 1
11 1
5 1
3 1
16 1
9 1
12 1
19 1
6 1
7 1
8 1
13 1
4 1
17 1
14 1
1 1
0 1
Name: count, dtype: int64
ddf.a.value_counts().head()
a
6 1
17 1
12 1
16 1
7 1
Name: count, dtype: int64
String Methods#
Like pandas, cuDF provides string processing methods in the str attribute of Series. Full documentation of string methods is a work in progress. Please see the cuDF API documentation for more information.
s = cudf.Series(["A", "B", "C", "Aaba", "Baca", None, "CABA", "dog", "cat"])
s.str.lower()
0 a
1 b
2 c
3 aaba
4 baca
5 <NA>
6 caba
7 dog
8 cat
dtype: object
ds = dask_cudf.from_cudf(s, npartitions=2)
ds.str.lower().head(n=4)
0 a
1 b
2 c
3 aaba
dtype: object
As well as simple manipulation, We can also match strings using regular expressions.
s.str.match("^[aAc].+")
0 False
1 False
2 False
3 True
4 False
5 <NA>
6 False
7 False
8 True
dtype: bool
ds.str.match("^[aAc].+").head()
0 False
1 False
2 False
3 True
4 False
dtype: bool
Concat#
Concatenating Series and DataFrames row-wise.
s = cudf.Series([1, 2, 3, None, 5])
cudf.concat([s, s])
0 1
1 2
2 3
3 <NA>
4 5
0 1
1 2
2 3
3 <NA>
4 5
dtype: int64
ds2 = dask_cudf.from_cudf(s, npartitions=2)
dask_cudf.concat([ds2, ds2]).head(n=3)
0 1
1 2
2 3
dtype: int64
Join#
Performing SQL style merges. Note that the dataframe order is not maintained, but may be restored post-merge by sorting by the index.
df_a = cudf.DataFrame()
df_a["key"] = ["a", "b", "c", "d", "e"]
df_a["vals_a"] = [float(i + 10) for i in range(5)]
df_b = cudf.DataFrame()
df_b["key"] = ["a", "c", "e"]
df_b["vals_b"] = [float(i + 100) for i in range(3)]
merged = df_a.merge(df_b, on=["key"], how="left")
merged
| key | vals_a | vals_b | |
|---|---|---|---|
| 0 | a | 10.0 | 100.0 |
| 1 | c | 12.0 | 101.0 |
| 2 | e | 14.0 | 102.0 |
| 3 | b | 11.0 | <NA> |
| 4 | d | 13.0 | <NA> |
ddf_a = dask_cudf.from_cudf(df_a, npartitions=2)
ddf_b = dask_cudf.from_cudf(df_b, npartitions=2)
merged = ddf_a.merge(ddf_b, on=["key"], how="left").head(n=4)
merged
| key | vals_a | vals_b | |
|---|---|---|---|
| 0 | e | 14.0 | 102.0 |
| 1 | c | 12.0 | 101.0 |
| 2 | d | 13.0 | <NA> |
| 3 | b | 11.0 | <NA> |
Grouping#
Like pandas, cuDF and Dask-cuDF support the Split-Apply-Combine groupby paradigm.
df["agg_col1"] = [1 if x % 2 == 0 else 0 for x in range(len(df))]
df["agg_col2"] = [1 if x % 3 == 0 else 0 for x in range(len(df))]
ddf = dask_cudf.from_cudf(df, npartitions=2)
Grouping and then applying the sum function to the grouped data.
df.groupby("agg_col1").sum()
| a | b | c | agg_col2 | |
|---|---|---|---|---|
| agg_col1 | ||||
| 1 | 90 | 100 | 90 | 4 |
| 0 | 100 | 90 | 100 | 3 |
ddf.groupby("agg_col1").sum().compute()
| a | b | c | agg_col2 | |
|---|---|---|---|---|
| agg_col1 | ||||
| 1 | 90 | 100 | 90 | 4 |
| 0 | 100 | 90 | 100 | 3 |
Grouping hierarchically then applying the sum function to grouped data.
df.groupby(["agg_col1", "agg_col2"]).sum()
| a | b | c | ||
|---|---|---|---|---|
| agg_col1 | agg_col2 | |||
| 1 | 1 | 36 | 40 | 36 |
| 0 | 54 | 60 | 54 | |
| 0 | 1 | 27 | 30 | 27 |
| 0 | 73 | 60 | 73 |
ddf.groupby(["agg_col1", "agg_col2"]).sum().compute()
| a | b | c | ||
|---|---|---|---|---|
| agg_col1 | agg_col2 | |||
| 1 | 1 | 36 | 40 | 36 |
| 0 | 0 | 73 | 60 | 73 |
| 1 | 27 | 30 | 27 | |
| 1 | 0 | 54 | 60 | 54 |
Grouping and applying statistical functions to specific columns, using agg.
df.groupby("agg_col1").agg({"a": "max", "b": "mean", "c": "sum"})
| a | b | c | |
|---|---|---|---|
| agg_col1 | |||
| 1 | 18 | 10.0 | 90 |
| 0 | 19 | 9.0 | 100 |
ddf.groupby("agg_col1").agg({"a": "max", "b": "mean", "c": "sum"}).compute()
| a | b | c | |
|---|---|---|---|
| agg_col1 | |||
| 1 | 18 | 10.0 | 90 |
| 0 | 19 | 9.0 | 100 |
Transpose#
Transposing a dataframe, using either the transpose method or T property. Currently, all columns must have the same type. Transposing is not currently implemented in Dask cuDF.
sample = cudf.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]})
sample
| a | b | |
|---|---|---|
| 0 | 1 | 4 |
| 1 | 2 | 5 |
| 2 | 3 | 6 |
sample.transpose()
| 0 | 1 | 2 | |
|---|---|---|---|
| a | 1 | 2 | 3 |
| b | 4 | 5 | 6 |
Time Series#
DataFrames supports datetime typed columns, which allow users to interact with and filter data based on specific timestamps.
import datetime as dt
date_df = cudf.DataFrame()
date_df["date"] = pd.date_range("11/20/2018", periods=72, freq="D")
date_df["value"] = cp.random.sample(len(date_df))
search_date = dt.datetime.strptime("2018-11-23", "%Y-%m-%d")
date_df.query("date <= @search_date")
| date | value | |
|---|---|---|
| 0 | 2018-11-20 | 0.986051 |
| 1 | 2018-11-21 | 0.232034 |
| 2 | 2018-11-22 | 0.397617 |
| 3 | 2018-11-23 | 0.103839 |
date_ddf = dask_cudf.from_cudf(date_df, npartitions=2)
date_ddf.query(
"date <= @search_date", local_dict={"search_date": search_date}
).compute()
| date | value | |
|---|---|---|
| 0 | 2018-11-20 | 0.986051 |
| 1 | 2018-11-21 | 0.232034 |
| 2 | 2018-11-22 | 0.397617 |
| 3 | 2018-11-23 | 0.103839 |
Categoricals#
DataFrames support categorical columns.
gdf = cudf.DataFrame(
{"id": [1, 2, 3, 4, 5, 6], "grade": ["a", "b", "b", "a", "a", "e"]}
)
gdf["grade"] = gdf["grade"].astype("category")
gdf
| id | grade | |
|---|---|---|
| 0 | 1 | a |
| 1 | 2 | b |
| 2 | 3 | b |
| 3 | 4 | a |
| 4 | 5 | a |
| 5 | 6 | e |
dgdf = dask_cudf.from_cudf(gdf, npartitions=2)
dgdf.head(n=3)
| id | grade | |
|---|---|---|
| 0 | 1 | a |
| 1 | 2 | b |
| 2 | 3 | b |
Accessing the categories of a column. Note that this is currently not supported in Dask-cuDF.
gdf.grade.cat.categories
Index(['a', 'b', 'e'], dtype='object')
Accessing the underlying code values of each categorical observation.
gdf.grade.cat.codes
0 0
1 1
2 1
3 0
4 0
5 2
dtype: uint8
dgdf.grade.cat.codes.compute()
0 0
1 1
2 1
3 0
4 0
5 2
dtype: uint8
Converting to Pandas#
Converting a cuDF and Dask-cuDF DataFrame to a pandas DataFrame.
df.head().to_pandas()
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
To convert the first few entries to pandas, we similarly call .head() on the Dask-cuDF dataframe to obtain a local cuDF dataframe, which we can then convert.
ddf.head().to_pandas()
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
In contrast, if we want to convert the entire frame, we need to call .compute() on ddf to get a local cuDF dataframe, and then call to_pandas(), followed by subsequent processing. This workflow is less recommended, since it both puts high memory pressure on a single GPU (the .compute() call) and does not take advantage of GPU acceleration for processing (the computation happens on in pandas).
ddf.compute().to_pandas().head()
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
Converting to Numpy#
Converting a cuDF or Dask-cuDF DataFrame to a numpy ndarray.
df.to_numpy()
array([[ 0, 19, 0, 1, 1],
[ 1, 18, 1, 0, 0],
[ 2, 17, 2, 1, 0],
[ 3, 16, 3, 0, 1],
[ 4, 15, 4, 1, 0],
[ 5, 14, 5, 0, 0],
[ 6, 13, 6, 1, 1],
[ 7, 12, 7, 0, 0],
[ 8, 11, 8, 1, 0],
[ 9, 10, 9, 0, 1],
[10, 9, 10, 1, 0],
[11, 8, 11, 0, 0],
[12, 7, 12, 1, 1],
[13, 6, 13, 0, 0],
[14, 5, 14, 1, 0],
[15, 4, 15, 0, 1],
[16, 3, 16, 1, 0],
[17, 2, 17, 0, 0],
[18, 1, 18, 1, 1],
[19, 0, 19, 0, 0]])
ddf.compute().to_numpy()
array([[ 0, 19, 0, 1, 1],
[ 1, 18, 1, 0, 0],
[ 2, 17, 2, 1, 0],
[ 3, 16, 3, 0, 1],
[ 4, 15, 4, 1, 0],
[ 5, 14, 5, 0, 0],
[ 6, 13, 6, 1, 1],
[ 7, 12, 7, 0, 0],
[ 8, 11, 8, 1, 0],
[ 9, 10, 9, 0, 1],
[10, 9, 10, 1, 0],
[11, 8, 11, 0, 0],
[12, 7, 12, 1, 1],
[13, 6, 13, 0, 0],
[14, 5, 14, 1, 0],
[15, 4, 15, 0, 1],
[16, 3, 16, 1, 0],
[17, 2, 17, 0, 0],
[18, 1, 18, 1, 1],
[19, 0, 19, 0, 0]])
Converting a cuDF or Dask-cuDF Series to a numpy ndarray.
df["a"].to_numpy()
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
17, 18, 19])
ddf["a"].compute().to_numpy()
array([ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
17, 18, 19])
Converting to Arrow#
Converting a cuDF or Dask-cuDF DataFrame to a PyArrow Table.
df.to_arrow()
pyarrow.Table
a: int64
b: int64
c: int64
agg_col1: int64
agg_col2: int64
----
a: [[0,1,2,3,4,...,15,16,17,18,19]]
b: [[19,18,17,16,15,...,4,3,2,1,0]]
c: [[0,1,2,3,4,...,15,16,17,18,19]]
agg_col1: [[1,0,1,0,1,...,0,1,0,1,0]]
agg_col2: [[1,0,0,1,0,...,1,0,0,1,0]]
ddf.head().to_arrow()
pyarrow.Table
a: int64
b: int64
c: int64
agg_col1: int64
agg_col2: int64
----
a: [[0,1,2,3,4]]
b: [[19,18,17,16,15]]
c: [[0,1,2,3,4]]
agg_col1: [[1,0,1,0,1]]
agg_col2: [[1,0,0,1,0]]
Reading/Writing CSV Files#
Writing to a CSV file.
if not os.path.exists("example_output"):
os.mkdir("example_output")
df.to_csv("example_output/foo.csv", index=False)
ddf.compute().to_csv("example_output/foo_dask.csv", index=False)
Reading from a csv file.
df = cudf.read_csv("example_output/foo.csv")
df
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
| 5 | 5 | 14 | 5 | 0 | 0 |
| 6 | 6 | 13 | 6 | 1 | 1 |
| 7 | 7 | 12 | 7 | 0 | 0 |
| 8 | 8 | 11 | 8 | 1 | 0 |
| 9 | 9 | 10 | 9 | 0 | 1 |
| 10 | 10 | 9 | 10 | 1 | 0 |
| 11 | 11 | 8 | 11 | 0 | 0 |
| 12 | 12 | 7 | 12 | 1 | 1 |
| 13 | 13 | 6 | 13 | 0 | 0 |
| 14 | 14 | 5 | 14 | 1 | 0 |
| 15 | 15 | 4 | 15 | 0 | 1 |
| 16 | 16 | 3 | 16 | 1 | 0 |
| 17 | 17 | 2 | 17 | 0 | 0 |
| 18 | 18 | 1 | 18 | 1 | 1 |
| 19 | 19 | 0 | 19 | 0 | 0 |
Note that for the Dask-cuDF case, we use dask_cudf.read_csv in preference to dask_cudf.from_cudf(cudf.read_csv) since the former can parallelize across multiple GPUs and handle larger CSV files that would fit in memory on a single GPU.
ddf = dask_cudf.read_csv("example_output/foo_dask.csv")
ddf.head()
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
Reading all CSV files in a directory into a single dask_cudf.DataFrame, using the star wildcard.
ddf = dask_cudf.read_csv("example_output/*.csv")
ddf.head()
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
Reading/Writing Parquet Files#
Writing to parquet files with cuDF’s GPU-accelerated parquet writer
df.to_parquet("example_output/temp_parquet")
Reading parquet files with cuDF’s GPU-accelerated parquet reader.
df = cudf.read_parquet("example_output/temp_parquet")
df
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
| 5 | 5 | 14 | 5 | 0 | 0 |
| 6 | 6 | 13 | 6 | 1 | 1 |
| 7 | 7 | 12 | 7 | 0 | 0 |
| 8 | 8 | 11 | 8 | 1 | 0 |
| 9 | 9 | 10 | 9 | 0 | 1 |
| 10 | 10 | 9 | 10 | 1 | 0 |
| 11 | 11 | 8 | 11 | 0 | 0 |
| 12 | 12 | 7 | 12 | 1 | 1 |
| 13 | 13 | 6 | 13 | 0 | 0 |
| 14 | 14 | 5 | 14 | 1 | 0 |
| 15 | 15 | 4 | 15 | 0 | 1 |
| 16 | 16 | 3 | 16 | 1 | 0 |
| 17 | 17 | 2 | 17 | 0 | 0 |
| 18 | 18 | 1 | 18 | 1 | 1 |
| 19 | 19 | 0 | 19 | 0 | 0 |
Writing to parquet files from a dask_cudf.DataFrame using cuDF’s parquet writer under the hood.
ddf.to_parquet("example_output/ddf_parquet_files")
Reading/Writing ORC Files#
Writing ORC files.
df.to_orc("example_output/temp_orc")
And reading
df2 = cudf.read_orc("example_output/temp_orc")
df2
| a | b | c | agg_col1 | agg_col2 | |
|---|---|---|---|---|---|
| 0 | 0 | 19 | 0 | 1 | 1 |
| 1 | 1 | 18 | 1 | 0 | 0 |
| 2 | 2 | 17 | 2 | 1 | 0 |
| 3 | 3 | 16 | 3 | 0 | 1 |
| 4 | 4 | 15 | 4 | 1 | 0 |
| 5 | 5 | 14 | 5 | 0 | 0 |
| 6 | 6 | 13 | 6 | 1 | 1 |
| 7 | 7 | 12 | 7 | 0 | 0 |
| 8 | 8 | 11 | 8 | 1 | 0 |
| 9 | 9 | 10 | 9 | 0 | 1 |
| 10 | 10 | 9 | 10 | 1 | 0 |
| 11 | 11 | 8 | 11 | 0 | 0 |
| 12 | 12 | 7 | 12 | 1 | 1 |
| 13 | 13 | 6 | 13 | 0 | 0 |
| 14 | 14 | 5 | 14 | 1 | 0 |
| 15 | 15 | 4 | 15 | 0 | 1 |
| 16 | 16 | 3 | 16 | 1 | 0 |
| 17 | 17 | 2 | 17 | 0 | 0 |
| 18 | 18 | 1 | 18 | 1 | 1 |
| 19 | 19 | 0 | 19 | 0 | 0 |
Dask Performance Tips#
Like Apache Spark, Dask operations are lazy. Instead of being executed immediately, most operations are added to a task graph and the actual evaluation is delayed until the result is needed.
Sometimes, though, we want to force the execution of operations. Calling persist on a Dask collection fully computes it (or actively computes it in the background), persisting the result into memory. When we’re using distributed systems, we may want to wait until persist is finished before beginning any downstream operations. We can enforce this contract by using wait. Wrapping an operation with wait will ensure it doesn’t begin executing until all necessary upstream operations have finished.
The snippets below provide basic examples, using LocalCUDACluster to create one dask-worker per GPU on the local machine. For more detailed information about persist and wait, please see the Dask documentation for persist and wait. Wait relies on the concept of Futures, which is beyond the scope of this tutorial. For more information on Futures, see the Dask Futures documentation. For more information about multi-GPU clusters, please see the dask-cuda library (documentation is in progress).
First, we set up a GPU cluster. With our client set up, Dask-cuDF computation will be distributed across the GPUs in the cluster.
import time
from dask.distributed import Client, wait
from dask_cuda import LocalCUDACluster
cluster = LocalCUDACluster()
client = Client(cluster)
Persisting Data#
Next, we create our Dask-cuDF DataFrame and apply a transformation, storing the result as a new column.
nrows = 10000000
df2 = cudf.DataFrame({"a": cp.arange(nrows), "b": cp.arange(nrows)})
ddf2 = dask_cudf.from_cudf(df2, npartitions=16)
ddf2["c"] = ddf2["a"] + 5
ddf2
| a | b | c | |
|---|---|---|---|
| npartitions=16 | |||
| 0 | int64 | int64 | int64 |
| 625000 | ... | ... | ... |
| ... | ... | ... | ... |
| 9375000 | ... | ... | ... |
| 9999999 | ... | ... | ... |
!nvidia-smi
Mon Dec 16 06:24:42 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15 Driver Version: 550.54.15 CUDA Version: 12.5 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla V100-PCIE-32GB Off | 00000000:85:00.0 Off | 0 |
| N/A 24C P0 36W / 250W | 645MiB / 32768MiB | 4% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+
Because Dask is lazy, the computation has not yet occurred. We can see that there are sixty-four tasks in the task graph and we’re using about 330 MB of device memory on each GPU. We can force computation by using persist. By forcing execution, the result is now explicitly in memory and our task graph only contains one task per partition (the baseline).
ddf2 = ddf2.persist()
ddf2
/opt/conda/envs/docs/lib/python3.11/site-packages/distributed/client.py:3371: UserWarning: Sending large graph of size 152.61 MiB.
This may cause some slowdown.
Consider loading the data with Dask directly
or using futures or delayed objects to embed the data into the graph without repetition.
See also https://docs.dask.org/en/stable/best-practices.html#load-data-with-dask for more information.
warnings.warn(
| a | b | c | |
|---|---|---|---|
| npartitions=16 | |||
| 0 | int64 | int64 | int64 |
| 625000 | ... | ... | ... |
| ... | ... | ... | ... |
| 9375000 | ... | ... | ... |
| 9999999 | ... | ... | ... |
# Sleep to ensure the persist finishes and shows in the memory usage
!sleep 5; nvidia-smi
Mon Dec 16 06:24:48 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15 Driver Version: 550.54.15 CUDA Version: 12.5 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 Tesla V100-PCIE-32GB Off | 00000000:85:00.0 Off | 0 |
| N/A 24C P0 36W / 250W | 1435MiB / 32768MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+
Because we forced computation, we now have a larger object in distributed GPU memory. Note that actual numbers will differ between systems (for example depending on how many devices are available).
Wait#
Depending on our workflow or distributed computing setup, we may want to wait until all upstream tasks have finished before proceeding with a specific function. This section shows an example of this behavior, adapted from the Dask documentation.
First, we create a new Dask DataFrame and define a function that we’ll map to every partition in the dataframe.
import random
nrows = 10000000
df1 = cudf.DataFrame({"a": cp.arange(nrows), "b": cp.arange(nrows)})
ddf1 = dask_cudf.from_cudf(df1, npartitions=100)
def func(df):
time.sleep(random.randint(1, 10))
return (df + 5) * 3 - 11
This function will do a basic transformation of every column in the dataframe, but the time spent in the function will vary due to the time.sleep statement randomly adding 1-10 seconds of time. We’ll run this on every partition of our dataframe using map_partitions, which adds the task to our task-graph, and store the result. We can then call persist to force execution.
results_ddf = ddf2.map_partitions(func)
results_ddf = results_ddf.persist()
However, some partitions will be done much sooner than others. If we had downstream processes that should wait for all partitions to be completed, we can enforce that behavior using wait.
wait(results_ddf)
DoneAndNotDoneFutures(done={<Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 10)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 12)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 2)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 11)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 15)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 0)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 4)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 5)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 13)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 3)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 1)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 9)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 7)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 14)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 8)>, <Future: finished, type: cudf.core.dataframe.DataFrame, key: ('func-90462946738802efa5b18a4f851570c8', 6)>}, not_done=set())
With wait completed, we can safely proceed on in our workflow.