You can run this notebook in a live session Binder or view it on Github.

Dask logo\

Dask DataFrames

We finished Chapter 1 by building a parallel dataframe computation over a directory of CSV files using dask.delayed. In this section we use dask.dataframe to automatically build similiar computations, for the common case of tabular computations. Dask dataframes look and feel like Pandas dataframes but they run on the same infrastructure that powers dask.delayed.

In this notebook we use the same airline data as before, but now rather than write for-loops we let dask.dataframe construct our computations for us. The dask.dataframe.read_csv function can take a globstring like "data/nycflights/*.csv" and build parallel computations on all of our data at once.

When to use dask.dataframe

Pandas is great for tabular datasets that fit in memory. Dask becomes useful when the dataset you want to analyze is larger than your machine’s RAM. The demo dataset we’re working with is only about 200MB, so that you can download it in a reasonable time, but dask.dataframe will scale to datasets much larger than memory.

467df822860a4c34932a5118fdf2ec55

The dask.dataframe module implements a blocked parallel DataFrame object that mimics a large subset of the Pandas DataFrame. One Dask DataFrame is comprised of many in-memory pandas DataFrames separated along the index. One operation on a Dask DataFrame triggers many pandas operations on the constituent pandas DataFrames in a way that is mindful of potential parallelism and memory constraints.

Related Documentation

Main Take-aways

  1. Dask DataFrame should be familiar to Pandas users

  2. The partitioning of dataframes is important for efficient execution

Create data

[1]:
%run prep.py -d flights

Setup

[2]:
from dask.distributed import Client

client = Client(n_workers=4)

We create artifical data.

[3]:
from prep import accounts_csvs
accounts_csvs()

import os
import dask
filename = os.path.join('data', 'accounts.*.csv')
filename
[3]:
'data/accounts.*.csv'

Filename includes a glob pattern *, so all files in the path matching that pattern will be read into the same Dask DataFrame.

[4]:
import dask.dataframe as dd
df = dd.read_csv(filename)
df.head()
[4]:
id names amount
0 85 George -521
1 36 Tim 134
2 91 Charlie 9747
3 34 Alice 7376
4 48 Ingrid 937
[5]:
# load and count number of rows
len(df)
[5]:
30000

What happened here? - Dask investigated the input path and found that there are three matching files - a set of jobs was intelligently created for each chunk - one per original CSV file in this case - each file was loaded into a pandas dataframe, had len() applied to it - the subtotals were combined to give you the final grand total.

Real Data

Lets try this with an extract of flights in the USA across several years. This data is specific to flights out of the three airports in the New York City area.

[6]:
df = dd.read_csv(os.path.join('data', 'nycflights', '*.csv'),
                 parse_dates={'Date': [0, 1, 2]})

Notice that the respresentation of the dataframe object contains no data - Dask has just done enough to read the start of the first file, and infer the column names and dtypes.

[7]:
df
[7]:
Dask DataFrame Structure:
Date DayOfWeek DepTime CRSDepTime ArrTime CRSArrTime UniqueCarrier FlightNum TailNum ActualElapsedTime CRSElapsedTime AirTime ArrDelay DepDelay Origin Dest Distance TaxiIn TaxiOut Cancelled Diverted
npartitions=10
datetime64[ns] int64 float64 int64 float64 int64 object int64 float64 float64 int64 float64 float64 float64 object object float64 float64 float64 int64 int64
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
Dask Name: read-csv, 10 tasks

We can view the start and end of the data

[8]:
df.head()
[8]:
Date DayOfWeek DepTime CRSDepTime ArrTime CRSArrTime UniqueCarrier FlightNum TailNum ActualElapsedTime ... AirTime ArrDelay DepDelay Origin Dest Distance TaxiIn TaxiOut Cancelled Diverted
0 1990-01-01 1 1621.0 1540 1747.0 1701 US 33 NaN 86.0 ... NaN 46.0 41.0 EWR PIT 319.0 NaN NaN 0 0
1 1990-01-02 2 1547.0 1540 1700.0 1701 US 33 NaN 73.0 ... NaN -1.0 7.0 EWR PIT 319.0 NaN NaN 0 0
2 1990-01-03 3 1546.0 1540 1710.0 1701 US 33 NaN 84.0 ... NaN 9.0 6.0 EWR PIT 319.0 NaN NaN 0 0
3 1990-01-04 4 1542.0 1540 1710.0 1701 US 33 NaN 88.0 ... NaN 9.0 2.0 EWR PIT 319.0 NaN NaN 0 0
4 1990-01-05 5 1549.0 1540 1706.0 1701 US 33 NaN 77.0 ... NaN 5.0 9.0 EWR PIT 319.0 NaN NaN 0 0

5 rows × 21 columns

[9]:
df.tail()  # this fails
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-9-430ef93b601c> in <module>
----> 1 df.tail()  # this fails

~/miniconda/envs/test/lib/python3.8/site-packages/dask/dataframe/core.py in tail(self, n, compute)
   1051
   1052         if compute:
-> 1053             result = result.compute()
   1054         return result
   1055

~/miniconda/envs/test/lib/python3.8/site-packages/dask/base.py in compute(self, **kwargs)
    164         dask.base.compute
    165         """
--> 166         (result,) = compute(self, traverse=False, **kwargs)
    167         return result
    168

~/miniconda/envs/test/lib/python3.8/site-packages/dask/base.py in compute(*args, **kwargs)
    442         postcomputes.append(x.__dask_postcompute__())
    443
--> 444     results = schedule(dsk, keys, **kwargs)
    445     return repack([f(r, *a) for r, (f, a) in zip(results, postcomputes)])
    446

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in get(self, dsk, keys, restrictions, loose_restrictions, resources, sync, asynchronous, direct, retries, priority, fifo_timeout, actors, **kwargs)
   2680                     should_rejoin = False
   2681             try:
-> 2682                 results = self.gather(packed, asynchronous=asynchronous, direct=direct)
   2683             finally:
   2684                 for f in futures.values():

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in gather(self, futures, errors, direct, asynchronous)
   1974             else:
   1975                 local_worker = None
-> 1976             return self.sync(
   1977                 self._gather,
   1978                 futures,

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in sync(self, func, asynchronous, callback_timeout, *args, **kwargs)
    829             return future
    830         else:
--> 831             return sync(
    832                 self.loop, func, *args, callback_timeout=callback_timeout, **kwargs
    833             )

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/utils.py in sync(loop, func, callback_timeout, *args, **kwargs)
    337     if error[0]:
    338         typ, exc, tb = error[0]
--> 339         raise exc.with_traceback(tb)
    340     else:
    341         return result[0]

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/utils.py in f()
    321             if callback_timeout is not None:
    322                 future = asyncio.wait_for(future, callback_timeout)
--> 323             result[0] = yield future
    324         except Exception as exc:
    325             error[0] = sys.exc_info()

~/miniconda/envs/test/lib/python3.8/site-packages/tornado/gen.py in run(self)
    733
    734                     try:
--> 735                         value = future.result()
    736                     except Exception:
    737                         exc_info = sys.exc_info()

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in _gather(self, futures, errors, direct, local_worker)
   1839                             exc = CancelledError(key)
   1840                         else:
-> 1841                             raise exception.with_traceback(traceback)
   1842                         raise exc
   1843                     if errors == "skip":

~/miniconda/envs/test/lib/python3.8/site-packages/dask/dataframe/io/csv.py in pandas_read_text()
    149     df = reader(bio, **kwargs)
    150     if dtypes:
--> 151         coerce_dtypes(df, dtypes)
    152
    153     if enforce and columns and (list(df.columns) != list(columns)):

~/miniconda/envs/test/lib/python3.8/site-packages/dask/dataframe/io/csv.py in coerce_dtypes()
    253             rule.join(filter(None, [dtype_msg, date_msg]))
    254         )
--> 255         raise ValueError(msg)
    256
    257

ValueError: Mismatched dtypes found in `pd.read_csv`/`pd.read_table`.

+----------------+---------+----------+
| Column         | Found   | Expected |
+----------------+---------+----------+
| CRSElapsedTime | float64 | int64    |
| TailNum        | object  | float64  |
+----------------+---------+----------+

The following columns also raised exceptions on conversion:

- TailNum
  ValueError("could not convert string to float: 'N54711'")

Usually this is due to dask's dtype inference failing, and
*may* be fixed by specifying dtypes manually by adding:

dtype={'CRSElapsedTime': 'float64',
       'TailNum': 'object'}

to the call to `read_csv`/`read_table`.

What just happened?

Unlike pandas.read_csv which reads in the entire file before inferring datatypes, dask.dataframe.read_csv only reads in a sample from the beginning of the file (or first file if using a glob). These inferred datatypes are then enforced when reading all partitions.

In this case, the datatypes inferred in the sample are incorrect. The first n rows have no value for CRSElapsedTime (which pandas infers as a float), and later on turn out to be strings (object dtype). Note that Dask gives an informative error message about the mismatch. When this happens you have a few options:

  • Specify dtypes directly using the dtype keyword. This is the recommended solution, as it’s the least error prone (better to be explicit than implicit) and also the most performant.

  • Increase the size of the sample keyword (in bytes)

  • Use assume_missing to make dask assume that columns inferred to be int (which don’t allow missing values) are actually floats (which do allow missing values). In our particular case this doesn’t apply.

In our case we’ll use the first option and directly specify the dtypes of the offending columns.

[10]:
df = dd.read_csv(os.path.join('data', 'nycflights', '*.csv'),
                 parse_dates={'Date': [0, 1, 2]},
                 dtype={'TailNum': str,
                        'CRSElapsedTime': float,
                        'Cancelled': bool})
[11]:
df.tail()  # now works
[11]:
Date DayOfWeek DepTime CRSDepTime ArrTime CRSArrTime UniqueCarrier FlightNum TailNum ActualElapsedTime ... AirTime ArrDelay DepDelay Origin Dest Distance TaxiIn TaxiOut Cancelled Diverted
994 1999-01-25 1 632.0 635 803.0 817 CO 437 N27213 91.0 ... 68.0 -14.0 -3.0 EWR RDU 416.0 4.0 19.0 False 0
995 1999-01-26 2 632.0 635 751.0 817 CO 437 N16217 79.0 ... 62.0 -26.0 -3.0 EWR RDU 416.0 3.0 14.0 False 0
996 1999-01-27 3 631.0 635 756.0 817 CO 437 N12216 85.0 ... 66.0 -21.0 -4.0 EWR RDU 416.0 4.0 15.0 False 0
997 1999-01-28 4 629.0 635 803.0 817 CO 437 N26210 94.0 ... 69.0 -14.0 -6.0 EWR RDU 416.0 5.0 20.0 False 0
998 1999-01-29 5 632.0 635 802.0 817 CO 437 N12225 90.0 ... 67.0 -15.0 -3.0 EWR RDU 416.0 5.0 18.0 False 0

5 rows × 21 columns

Computations with dask.dataframe

We compute the maximum of the DepDelay column. With just pandas, we would loop over each file to find the individual maximums, then find the final maximum over all the individual maximums

maxes = []
for fn in filenames:
    df = pd.read_csv(fn)
    maxes.append(df.DepDelay.max())

final_max = max(maxes)

We could wrap that pd.read_csv with dask.delayed so that it runs in parallel. Regardless, we’re still having to think about loops, intermediate results (one per file) and the final reduction (max of the intermediate maxes). This is just noise around the real task, which pandas solves with

df = pd.read_csv(filename, dtype=dtype)
df.DepDelay.max()

dask.dataframe lets us write pandas-like code, that operates on larger than memory datasets in parallel.

[12]:
%time df.DepDelay.max().compute()
CPU times: user 49 ms, sys: 1.9 ms, total: 50.9 ms
Wall time: 443 ms
[12]:
409.0

This writes the delayed computation for us and then runs it.

Some things to note:

  1. As with dask.delayed, we need to call .compute() when we’re done. Up until this point everything is lazy.

  2. Dask will delete intermediate results (like the full pandas dataframe for each file) as soon as possible.

    • This lets us handle datasets that are larger than memory

    • This means that repeated computations will have to load all of the data in each time (run the code above again, is it faster or slower than you would expect?)

As with Delayed objects, you can view the underlying task graph using the .visualize method:

[13]:
# notice the parallelism
df.DepDelay.max().visualize()
[13]:
_images/04_dataframe_25_0.png

Exercises

In this section we do a few dask.dataframe computations. If you are comfortable with Pandas then these should be familiar. You will have to think about when to call compute.

1.) How many rows are in our dataset?

If you aren’t familiar with pandas, how would you check how many records are in a list of tuples?

[14]:
# Your code here
[15]:
len(df)
[15]:
9990

2.) In total, how many non-canceled flights were taken?

With pandas, you would use boolean indexing.

[16]:
# Your code here
[17]:
len(df[~df.Cancelled])
[17]:
9383

3.) In total, how many non-cancelled flights were taken from each airport?

Hint: use `df.groupby <https://pandas.pydata.org/pandas-docs/stable/groupby.html>`__.

[18]:
# Your code here
[19]:
df[~df.Cancelled].groupby('Origin').Origin.count().compute()
[19]:
Origin
EWR    4132
JFK    1085
LGA    4166
Name: Origin, dtype: int64

4.) What was the average departure delay from each airport?

Note, this is the same computation you did in the previous notebook (is this approach faster or slower?)

[20]:
# Your code here
[21]:
df.groupby("Origin").DepDelay.mean().compute()
[21]:
Origin
EWR    12.500968
JFK    17.053456
LGA    10.169227
Name: DepDelay, dtype: float64

5.) What day of the week has the worst average departure delay?

[22]:
# Your code here
[23]:
df.groupby("DayOfWeek").DepDelay.mean().compute()
[23]:
DayOfWeek
1    10.677698
2     8.633310
3    14.208160
4    14.187853
5    15.209929
6     9.540307
7    10.609375
Name: DepDelay, dtype: float64

Sharing Intermediate Results

When computing all of the above, we sometimes did the same operation more than once. For most operations, dask.dataframe hashes the arguments, allowing duplicate computations to be shared, and only computed once.

For example, lets compute the mean and standard deviation for departure delay of all non-canceled flights. Since dask operations are lazy, those values aren’t the final results yet. They’re just the recipe required to get the result.

If we compute them with two calls to compute, there is no sharing of intermediate computations.

[24]:
non_cancelled = df[~df.Cancelled]
mean_delay = non_cancelled.DepDelay.mean()
std_delay = non_cancelled.DepDelay.std()
[25]:
%%time

mean_delay_res = mean_delay.compute()
std_delay_res = std_delay.compute()
CPU times: user 163 ms, sys: 14.3 ms, total: 177 ms
Wall time: 402 ms

But lets try by passing both to a single compute call.

[26]:
%%time

mean_delay_res, std_delay_res = dask.compute(mean_delay, std_delay)
CPU times: user 62.4 ms, sys: 1.3 ms, total: 63.7 ms
Wall time: 189 ms

Using dask.compute takes roughly 1/2 the time. This is because the task graphs for both results are merged when calling dask.compute, allowing shared operations to only be done once instead of twice. In particular, using dask.compute only does the following once:

  • the calls to read_csv

  • the filter (df[~df.Cancelled])

  • some of the necessary reductions (sum, count)

To see what the merged task graphs between multiple results look like (and what’s shared), you can use the dask.visualize function (we might want to use filename='graph.pdf' to zoom in on the graph better):

[27]:
dask.visualize(mean_delay, std_delay)
[27]:
_images/04_dataframe_47_0.png

How does this compare to Pandas?

Pandas is more mature and fully featured than dask.dataframe. If your data fits in memory then you should use Pandas. The dask.dataframe module gives you a limited pandas experience when you operate on datasets that don’t fit comfortably in memory.

During this tutorial we provide a small dataset consisting of a few CSV files. This dataset is 45MB on disk that expands to about 400MB in memory. This dataset is small enough that you would normally use Pandas.

We’ve chosen this size so that exercises finish quickly. Dask.dataframe only really becomes meaningful for problems significantly larger than this, when Pandas breaks with the dreaded

MemoryError:  ...

Furthermore, the distributed scheduler allows the same dataframe expressions to be executed across a cluster. To enable massive “big data” processing, one could execute data ingestion functions such as read_csv, where the data is held on storage accessible to every worker node (e.g., amazon’s S3), and because most operations begin by selecting only some columns, transforming and filtering the data, only relatively small amounts of data need to be communicated between the machines.

Dask.dataframe operations use pandas operations internally. Generally they run at about the same speed except in the following two cases:

  1. Dask introduces a bit of overhead, around 1ms per task. This is usually negligible.

  2. When Pandas releases the GIL (coming to groupby in the next version) dask.dataframe can call several pandas operations in parallel within a process, increasing speed somewhat proportional to the number of cores. For operations which don’t release the GIL, multiple processes would be needed to get the same speedup.

Dask DataFrame Data Model

For the most part, a Dask DataFrame feels like a pandas DataFrame. So far, the biggest difference we’ve seen is that Dask operations are lazy; they build up a task graph instead of executing immediately (more details coming in Schedulers). This lets Dask do operations in parallel and out of core.

In Dask Arrays, we saw that a dask.array was composed of many NumPy arrays, chunked along one or more dimensions. It’s similar for dask.dataframe: a Dask DataFrame is composed of many pandas DataFrames. For dask.dataframe the chunking happens only along the index.

c77f3b92e95847c78d5f508a1ac1ba93

We call each chunk a partition, and the upper / lower bounds are divisions. Dask can store information about the divisions. For now, partitions come up when you write custom functions to apply to Dask DataFrames

Converting CRSDepTime to a timestamp

This dataset stores timestamps as HHMM, which are read in as integers in read_csv:

[28]:
crs_dep_time = df.CRSDepTime.head(10)
crs_dep_time
[28]:
0    1540
1    1540
2    1540
3    1540
4    1540
5    1540
6    1540
7    1540
8    1540
9    1540
Name: CRSDepTime, dtype: int64

To convert these to timestamps of scheduled departure time, we need to convert these integers into pd.Timedelta objects, and then combine them with the Date column.

In pandas we’d do this using the pd.to_timedelta function, and a bit of arithmetic:

[29]:
import pandas as pd

# Get the first 10 dates to complement our `crs_dep_time`
date = df.Date.head(10)

# Get hours as an integer, convert to a timedelta
hours = crs_dep_time // 100
hours_timedelta = pd.to_timedelta(hours, unit='h')

# Get minutes as an integer, convert to a timedelta
minutes = crs_dep_time % 100
minutes_timedelta = pd.to_timedelta(minutes, unit='m')

# Apply the timedeltas to offset the dates by the departure time
departure_timestamp = date + hours_timedelta + minutes_timedelta
departure_timestamp
[29]:
0   1990-01-01 15:40:00
1   1990-01-02 15:40:00
2   1990-01-03 15:40:00
3   1990-01-04 15:40:00
4   1990-01-05 15:40:00
5   1990-01-06 15:40:00
6   1990-01-07 15:40:00
7   1990-01-08 15:40:00
8   1990-01-09 15:40:00
9   1990-01-10 15:40:00
dtype: datetime64[ns]

Custom code and Dask Dataframe

We could swap out pd.to_timedelta for dd.to_timedelta and do the same operations on the entire dask DataFrame. But let’s say that Dask hadn’t implemented a dd.to_timedelta that works on Dask DataFrames. What would you do then?

dask.dataframe provides a few methods to make applying custom functions to Dask DataFrames easier:

Here we’ll just be discussing map_partitions, which we can use to implement to_timedelta on our own:

[30]:
# Look at the docs for `map_partitions`

help(df.CRSDepTime.map_partitions)
Help on method map_partitions in module dask.dataframe.core:

map_partitions(func, *args, **kwargs) method of dask.dataframe.core.Series instance
    Apply Python function on each DataFrame partition.

    Note that the index and divisions are assumed to remain unchanged.

    Parameters
    ----------
    func : function
        Function applied to each partition.
    args, kwargs :
        Arguments and keywords to pass to the function. The partition will
        be the first argument, and these will be passed *after*. Arguments
        and keywords may contain ``Scalar``, ``Delayed`` or regular
        python objects. DataFrame-like args (both dask and pandas) will be
        repartitioned to align (if necessary) before applying the function.
    meta : pd.DataFrame, pd.Series, dict, iterable, tuple, optional
        An empty ``pd.DataFrame`` or ``pd.Series`` that matches the dtypes
        and column names of the output. This metadata is necessary for
        many algorithms in dask dataframe to work.  For ease of use, some
        alternative inputs are also available. Instead of a ``DataFrame``,
        a ``dict`` of ``{name: dtype}`` or iterable of ``(name, dtype)``
        can be provided (note that the order of the names should match the
        order of the columns). Instead of a series, a tuple of ``(name,
        dtype)`` can be used. If not provided, dask will try to infer the
        metadata. This may lead to unexpected results, so providing
        ``meta`` is recommended. For more information, see
        ``dask.dataframe.utils.make_meta``.

    Examples
    --------
    Given a DataFrame, Series, or Index, such as:

    >>> import dask.dataframe as dd
    >>> df = pd.DataFrame({'x': [1, 2, 3, 4, 5],
    ...                    'y': [1., 2., 3., 4., 5.]})
    >>> ddf = dd.from_pandas(df, npartitions=2)

    One can use ``map_partitions`` to apply a function on each partition.
    Extra arguments and keywords can optionally be provided, and will be
    passed to the function after the partition.

    Here we apply a function with arguments and keywords to a DataFrame,
    resulting in a Series:

    >>> def myadd(df, a, b=1):
    ...     return df.x + df.y + a + b
    >>> res = ddf.map_partitions(myadd, 1, b=2)
    >>> res.dtype
    dtype('float64')

    By default, dask tries to infer the output metadata by running your
    provided function on some fake data. This works well in many cases, but
    can sometimes be expensive, or even fail. To avoid this, you can
    manually specify the output metadata with the ``meta`` keyword. This
    can be specified in many forms, for more information see
    ``dask.dataframe.utils.make_meta``.

    Here we specify the output is a Series with no name, and dtype
    ``float64``:

    >>> res = ddf.map_partitions(myadd, 1, b=2, meta=(None, 'f8'))

    Here we map a function that takes in a DataFrame, and returns a
    DataFrame with a new column:

    >>> res = ddf.map_partitions(lambda df: df.assign(z=df.x * df.y))
    >>> res.dtypes
    x      int64
    y    float64
    z    float64
    dtype: object

    As before, the output metadata can also be specified manually. This
    time we pass in a ``dict``, as the output is a DataFrame:

    >>> res = ddf.map_partitions(lambda df: df.assign(z=df.x * df.y),
    ...                          meta={'x': 'i8', 'y': 'f8', 'z': 'f8'})

    In the case where the metadata doesn't change, you can also pass in
    the object itself directly:

    >>> res = ddf.map_partitions(lambda df: df.head(), meta=ddf)

    Also note that the index and divisions are assumed to remain unchanged.
    If the function you're mapping changes the index/divisions, you'll need
    to clear them afterwards:

    >>> ddf.map_partitions(func).clear_divisions()  # doctest: +SKIP

The basic idea is to apply a function that operates on a DataFrame to each partition. In this case, we’ll apply pd.to_timedelta.

[31]:
hours = df.CRSDepTime // 100
# hours_timedelta = pd.to_timedelta(hours, unit='h')
hours_timedelta = hours.map_partitions(pd.to_timedelta, unit='h')

minutes = df.CRSDepTime % 100
# minutes_timedelta = pd.to_timedelta(minutes, unit='m')
minutes_timedelta = minutes.map_partitions(pd.to_timedelta, unit='m')

departure_timestamp = df.Date + hours_timedelta + minutes_timedelta
[32]:
departure_timestamp
[32]:
Dask Series Structure:
npartitions=10
    datetime64[ns]
               ...
         ...
               ...
               ...
dtype: datetime64[ns]
Dask Name: add, 90 tasks
[33]:
departure_timestamp.head()
[33]:
0   1990-01-01 15:40:00
1   1990-01-02 15:40:00
2   1990-01-03 15:40:00
3   1990-01-04 15:40:00
4   1990-01-05 15:40:00
dtype: datetime64[ns]

Exercise: Rewrite above to use a single call to map_partitions

This will be slightly more efficient than two separate calls, as it reduces the number of tasks in the graph.

[34]:
def compute_departure_timestamp(df):
    pass  # TODO: implement this
[35]:
departure_timestamp = df.map_partitions(compute_departure_timestamp)

departure_timestamp.head()
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-35-b8c8b4861a84> in <module>
      1 departure_timestamp = df.map_partitions(compute_departure_timestamp)
      2
----> 3 departure_timestamp.head()

~/miniconda/envs/test/lib/python3.8/site-packages/dask/dataframe/core.py in head(self, n, npartitions, compute)
   1003             Whether to compute the result, default is True.
   1004         """
-> 1005         return self._head(n=n, npartitions=npartitions, compute=compute, safe=True)
   1006
   1007     def _head(self, n, npartitions, compute, safe):

~/miniconda/envs/test/lib/python3.8/site-packages/dask/dataframe/core.py in _head(self, n, npartitions, compute, safe)
   1036
   1037         if compute:
-> 1038             result = result.compute()
   1039         return result
   1040

~/miniconda/envs/test/lib/python3.8/site-packages/dask/base.py in compute(self, **kwargs)
    164         dask.base.compute
    165         """
--> 166         (result,) = compute(self, traverse=False, **kwargs)
    167         return result
    168

~/miniconda/envs/test/lib/python3.8/site-packages/dask/base.py in compute(*args, **kwargs)
    442         postcomputes.append(x.__dask_postcompute__())
    443
--> 444     results = schedule(dsk, keys, **kwargs)
    445     return repack([f(r, *a) for r, (f, a) in zip(results, postcomputes)])
    446

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in get(self, dsk, keys, restrictions, loose_restrictions, resources, sync, asynchronous, direct, retries, priority, fifo_timeout, actors, **kwargs)
   2680                     should_rejoin = False
   2681             try:
-> 2682                 results = self.gather(packed, asynchronous=asynchronous, direct=direct)
   2683             finally:
   2684                 for f in futures.values():

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in gather(self, futures, errors, direct, asynchronous)
   1974             else:
   1975                 local_worker = None
-> 1976             return self.sync(
   1977                 self._gather,
   1978                 futures,

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in sync(self, func, asynchronous, callback_timeout, *args, **kwargs)
    829             return future
    830         else:
--> 831             return sync(
    832                 self.loop, func, *args, callback_timeout=callback_timeout, **kwargs
    833             )

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/utils.py in sync(loop, func, callback_timeout, *args, **kwargs)
    337     if error[0]:
    338         typ, exc, tb = error[0]
--> 339         raise exc.with_traceback(tb)
    340     else:
    341         return result[0]

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/utils.py in f()
    321             if callback_timeout is not None:
    322                 future = asyncio.wait_for(future, callback_timeout)
--> 323             result[0] = yield future
    324         except Exception as exc:
    325             error[0] = sys.exc_info()

~/miniconda/envs/test/lib/python3.8/site-packages/tornado/gen.py in run(self)
    733
    734                     try:
--> 735                         value = future.result()
    736                     except Exception:
    737                         exc_info = sys.exc_info()

~/miniconda/envs/test/lib/python3.8/site-packages/distributed/client.py in _gather(self, futures, errors, direct, local_worker)
   1839                             exc = CancelledError(key)
   1840                         else:
-> 1841                             raise exception.with_traceback(traceback)
   1842                         raise exc
   1843                     if errors == "skip":

~/miniconda/envs/test/lib/python3.8/site-packages/dask/dataframe/core.py in safe_head()
   6169
   6170 def safe_head(df, n):
-> 6171     r = M.head(df, n)
   6172     if len(r) != n:
   6173         msg = (

~/miniconda/envs/test/lib/python3.8/site-packages/dask/utils.py in __call__()
    893
    894     def __call__(self, obj, *args, **kwargs):
--> 895         return getattr(obj, self.method)(*args, **kwargs)
    896
    897     def __reduce__(self):

AttributeError: 'NoneType' object has no attribute 'head'
[36]:
def compute_departure_timestamp(df):
    hours = df.CRSDepTime // 100
    hours_timedelta = pd.to_timedelta(hours, unit='h')

    minutes = df.CRSDepTime % 100
    minutes_timedelta = pd.to_timedelta(minutes, unit='m')

    return df.Date + hours_timedelta + minutes_timedelta

departure_timestamp = df.map_partitions(compute_departure_timestamp)
departure_timestamp.head()
[36]:
0   1990-01-01 15:40:00
1   1990-01-02 15:40:00
2   1990-01-03 15:40:00
3   1990-01-04 15:40:00
4   1990-01-05 15:40:00
dtype: datetime64[ns]

Limitations

What doesn’t work?

Dask.dataframe only covers a small but well-used portion of the Pandas API. This limitation is for two reasons:

  1. The Pandas API is huge

  2. Some operations are genuinely hard to do in parallel (e.g. sort)

Additionally, some important operations like set_index work, but are slower than in Pandas because they include substantial shuffling of data, and may write out to disk.