Install¶

If you don't have ibis installed, you can install it from:

PyPI¶

python -m pip install 'ibis-framework[duckdb]'

conda-forge¶

conda install -c conda-forge ibis-framework

You can check out the install page for more detailed instructions.

Ibis for pandas Users¶

Much of the syntax and many of the operations in Ibis are inspired by the pandas DataFrame but the primary domain of Ibis is SQL so there are some differences in how they operate.

One primary difference between Ibis tables and pandas DataFrames are that many of the pandas DataFrame operations do in-place operations (they are "mutable"), whereas Ibis table operations always return a new table expression ("immutable").

Another difference is that Ibis table expressions are lazy, meaning that as you build up a table expression, no computation is actually performed until you call an action method such as execute. Only then does Ibis compile the table expression into SQL and send it to the backend. (Note that we'll be using Ibis' interactive mode to automatically execute queries at the end of each cell in this notebook. If you are using similar code in a program, you will have to add .execute() to each operation that you want to evaluate.)

In [1]:

import ibis
import pandas as pd

ibis.options.interactive = True

import ibis import pandas as pd ibis.options.interactive = True

We'll be using the DuckDB backend in Ibis in the examples below. First we'll create a simple DataFrame.

In [2]:

df = pd.DataFrame(
    [['a', 1, 2], ['b', 3, 4]],
    columns=['one', 'two', 'three'],
    index=[5, 6],
)
df

df = pd.DataFrame( [['a', 1, 2], ['b', 3, 4]], columns=['one', 'two', 'three'], index=[5, 6], ) df

Out[2]:

	one	two	three
5	a	1	2
6	b	3	4

Now we can create an Ibis table from the above DataFrame.

Note that the index from the pandas DataFrame is dropped. Ibis has no notion of an index: If you want to use the index, you will need to turn it into a column.

In [3]:

t = ibis.memtable(df, name="t")
t

t = ibis.memtable(df, name="t") t

Out[3]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ one    ┃ two   ┃ three ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │
├────────┼───────┼───────┤
│ a      │     1 │     2 │
│ b      │     3 │     4 │
└────────┴───────┴───────┘

Data types¶

The data types of columns in pandas are accessed using the dtypes attribute. This returns a Series object.

In [4]:

df.dtypes

df.dtypes

Out[4]:

one      object
two       int64
three     int64
dtype: object

In Ibis, you use the schema method which returns an ibis.Schema object.

In [5]:

t.schema()

t.schema()

Out[5]:

ibis.Schema {
  one    string
  two    int64
  three  int64
}

It is possible to convert the schema information to pandas data types using the to_pandas method, if needed.

In [6]:

t.schema().to_pandas()

t.schema().to_pandas()

Out[6]:

[('one', dtype('O')), ('two', dtype('int64')), ('three', dtype('int64'))]

Table layout¶

In pandas, the layout of the table is contained in the shape attribute which contains the number of rows and number of columns in a tuple. The number of columns in an Ibis table can be gotten from the length of the schema.

In [7]:

len(t.schema())

len(t.schema())

Out[7]:

3

To get the number of rows of a table, you use the count method.

In [8]:

t.count()

t.count()

Out[8]:

2

To mimic pandas' behavior, you would use the following code. Note that you need to use the execute method after count to evaluate the expression returned by count.

In [9]:

(t.count().execute(), len(t.schema()))

(t.count().execute(), len(t.schema()))

Out[9]:

(2, 3)

In [10]:

df.shape

df.shape

Out[10]:

(2, 3)

Subsetting columns¶

Selecting columns is very similar to in pandas. In fact, you can use the same syntax.

In [11]:

t[['one', 'two']]

t[['one', 'two']]

Out[11]:

┏━━━━━━━━┳━━━━━━━┓
┃ one    ┃ two   ┃
┡━━━━━━━━╇━━━━━━━┩
│ string │ int64 │
├────────┼───────┤
│ a      │     1 │
│ b      │     3 │
└────────┴───────┘

However, since row-level indexing is not supported in Ibis, the inner list is not necessary.

In [12]:

t['one', 'two']

t['one', 'two']

Out[12]:

┏━━━━━━━━┳━━━━━━━┓
┃ one    ┃ two   ┃
┡━━━━━━━━╇━━━━━━━┩
│ string │ int64 │
├────────┼───────┤
│ a      │     1 │
│ b      │     3 │
└────────┴───────┘

Selecting columns¶

Selecting columns is done using the same syntax as in pandas DataFrames. You can use either the indexing syntax or attribute syntax.

In [13]:

t['one']

t['one']

Out[13]:

┏━━━━━━━━┓
┃ one    ┃
┡━━━━━━━━┩
│ string │
├────────┤
│ a      │
│ b      │
└────────┘

or:

In [14]:

t.one

t.one

Out[14]:

┏━━━━━━━━┓
┃ one    ┃
┡━━━━━━━━┩
│ string │
├────────┤
│ a      │
│ b      │
└────────┘

Adding, removing, and modifying columns¶

Modifying the columns of an Ibis table is a bit different than doing the same operations in a pandas DataFrame. This is primarily due to the fact that in-place operations are not supported on Ibis tables. Each time you do a column modification to a table, a new table expression is returned.

Adding columns¶

Adding columns is done through the mutate method.

In [15]:

mutated = t.mutate(new_col=t.three * 2)
mutated

mutated = t.mutate(new_col=t.three * 2) mutated

Out[15]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┳━━━━━━━━━┓
┃ one    ┃ two   ┃ three ┃ new_col ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━╇━━━━━━━━━┩
│ string │ int64 │ int64 │ int64   │
├────────┼───────┼───────┼─────────┤
│ a      │     1 │     2 │       4 │
│ b      │     3 │     4 │       8 │
└────────┴───────┴───────┴─────────┘

Notice that the original table object remains unchanged. Only the mutated object that was returned contains the new column.

In [16]:

t

t

Out[16]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ one    ┃ two   ┃ three ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │
├────────┼───────┼───────┤
│ a      │     1 │     2 │
│ b      │     3 │     4 │
└────────┴───────┴───────┘

It is also possible to create a column in isolation. This is similar to a Series in pandas. Note that the name of the column by default is a representation of the expression:

In [17]:

unnamed = t.three * 2
unnamed

unnamed = t.three * 2 unnamed

Out[17]:

┏━━━━━━━━━━━━━━━━━━━━┓
┃ Multiply(three, 2) ┃
┡━━━━━━━━━━━━━━━━━━━━┩
│ int64              │
├────────────────────┤
│                  4 │
│                  8 │
└────────────────────┘

To get a version with a specific name, you can use the name method:

In [18]:

new_col = unnamed.name("new_col")
new_col

new_col = unnamed.name("new_col") new_col

Out[18]:

┏━━━━━━━━━┓
┃ new_col ┃
┡━━━━━━━━━┩
│ int64   │
├─────────┤
│       4 │
│       8 │
└─────────┘

You can then add this column to the table using a projection.

In [19]:

proj = t['one', 'two', new_col]
proj

proj = t['one', 'two', new_col] proj

Out[19]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━━━┓
┃ one    ┃ two   ┃ new_col ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━━━┩
│ string │ int64 │ int64   │
├────────┼───────┼─────────┤
│ a      │     1 │       4 │
│ b      │     3 │       8 │
└────────┴───────┴─────────┘

Removing columns¶

Removing a column is done using the drop method.

In [20]:

t.columns

t.columns

Out[20]:

['one', 'two', 'three']

In [21]:

subset = t.drop('one', 'two')
subset.columns

subset = t.drop('one', 'two') subset.columns

Out[21]:

['three']

It is also possible to drop columns by selecting the columns you want to remain.

In [22]:

subset = t['two', 'three']
subset.columns

subset = t['two', 'three'] subset.columns

Out[22]:

['two', 'three']

Modifying columns¶

Replacing existing columns is done using the mutate method just like adding columns. You simply add a column of the same name to replace it.

In [23]:

t

t

Out[23]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ one    ┃ two   ┃ three ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │
├────────┼───────┼───────┤
│ a      │     1 │     2 │
│ b      │     3 │     4 │
└────────┴───────┴───────┘

In [24]:

mutated = t.mutate(two=t.two * 2)
mutated

mutated = t.mutate(two=t.two * 2) mutated

Out[24]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ one    ┃ two   ┃ three ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │
├────────┼───────┼───────┤
│ a      │     2 │     2 │
│ b      │     6 │     4 │
└────────┴───────┴───────┘

Renaming columns¶

In addition to replacing columns, you can simply rename them as well. This is done with the relabel method which takes a dictionary containing the name mappings.

In [25]:

relabeled = t.relabel(
    dict(
        one='a',
        two='b',
    )
)
relabeled

relabeled = t.relabel( dict( one='a', two='b', ) ) relabeled

Out[25]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ a      ┃ b     ┃ three ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │
├────────┼───────┼───────┤
│ a      │     1 │     2 │
│ b      │     3 │     4 │
└────────┴───────┴───────┘

Selecting rows¶

There are several methods that can be used to select rows of data in various ways. These are described in the sections below. We'll use the Palmer Penguins$^1$ dataset to investigate! Ibis has several built-in example datasets that you can access using the ibis.examples module.

$^1$: Horst AM, Hill AP, Gorman KB (2020). palmerpenguins: Palmer Archipelago (Antarctica) penguin data. R package version 0.1.0. https://allisonhorst.github.io/palmerpenguins/. doi: 10.5281/zenodo.3960218.

In [26]:

penguins = ibis.examples.penguins.fetch()

penguins = ibis.examples.penguins.fetch()

Downloading file 'penguins.csv.gz' from 'https://storage.googleapis.com/ibis-examples/data/penguins.csv.gz' to '/home/runner/.cache/ibis-framework/5.1.0'.
100%|█████████████████████████████████████| 2.96k/2.96k [00:00<00:00, 2.70MB/s]

In [27]:

penguins

penguins

Out[27]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.5 │          17.4 │               186 │        3800 │ female │  2007 │
│ Adelie  │ Torgersen │           40.3 │          18.0 │               195 │        3250 │ female │  2007 │
│ Adelie  │ Torgersen │            nan │           nan │              NULL │        NULL │ NULL   │  2007 │
│ Adelie  │ Torgersen │           36.7 │          19.3 │               193 │        3450 │ female │  2007 │
│ Adelie  │ Torgersen │           39.3 │          20.6 │               190 │        3650 │ male   │  2007 │
│ Adelie  │ Torgersen │           38.9 │          17.8 │               181 │        3625 │ female │  2007 │
│ Adelie  │ Torgersen │           39.2 │          19.6 │               195 │        4675 │ male   │  2007 │
│ Adelie  │ Torgersen │           34.1 │          18.1 │               193 │        3475 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           42.0 │          20.2 │               190 │        4250 │ NULL   │  2007 │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Head, tail and limit¶

The head method works the same ways as in pandas. Note that some Ibis backends may not have an inherent ordering of their rows and using head may not return deterministic results. In those cases, you can use sorting before calling head to ensure a stable result.

In [28]:

penguins.head(5)

penguins.head(5)

Out[28]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.5 │          17.4 │               186 │        3800 │ female │  2007 │
│ Adelie  │ Torgersen │           40.3 │          18.0 │               195 │        3250 │ female │  2007 │
│ Adelie  │ Torgersen │            nan │           nan │              NULL │        NULL │ NULL   │  2007 │
│ Adelie  │ Torgersen │           36.7 │          19.3 │               193 │        3450 │ female │  2007 │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

However, the tail method is not implemented since it is not supported in all databases. It is possible to emulate the tail method if you use sorting in your table to do a reverse sort then use the head method to retrieve the "top" rows.

Another way to limit the number of retrieved rows is using the limit method. The following will return the same result as head(5). This is often used in conjunction with other filtering techniques that we will cover later.

In [29]:

penguins.limit(5)

penguins.limit(5)

Out[29]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.5 │          17.4 │               186 │        3800 │ female │  2007 │
│ Adelie  │ Torgersen │           40.3 │          18.0 │               195 │        3250 │ female │  2007 │
│ Adelie  │ Torgersen │            nan │           nan │              NULL │        NULL │ NULL   │  2007 │
│ Adelie  │ Torgersen │           36.7 │          19.3 │               193 │        3450 │ female │  2007 │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Filtering rows¶

In addition to simply limiting the number of rows that are returned, it is possible to filter the rows using expressions. Expressions are constructed very similarly to the way they are in pandas. Ibis expressions are constructed from operations on columns in a table which return a boolean result. This result is then used to filter the table.

In [30]:

expr = penguins.bill_length_mm > 37.0
expr

expr = penguins.bill_length_mm > 37.0 expr

Out[30]:

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Greater(bill_length_mm, 37.0) ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ boolean                       │
├───────────────────────────────┤
│ True                          │
│ True                          │
│ True                          │
│ NULL                          │
│ False                         │
│ True                          │
│ True                          │
│ True                          │
│ False                         │
│ True                          │
│ …                             │
└───────────────────────────────┘

We can evaluate the value counts to see how many rows we will expect to get back after filtering.

In [31]:

expr.value_counts()

expr.value_counts()

Out[31]:

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Greater(bill_length_mm, 37.0) ┃ Greater(bill_length_mm, 37.0)_count ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ boolean                       │ int64                               │
├───────────────────────────────┼─────────────────────────────────────┤
│ True                          │                                 300 │
│ NULL                          │                                   2 │
│ False                         │                                  42 │
└───────────────────────────────┴─────────────────────────────────────┘

Now we apply the filter to the table. Since there are 6 True values in the expression, we should get 6 rows back.

In [32]:

filtered = penguins[expr]
filtered

filtered = penguins[expr] filtered

Out[32]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.5 │          17.4 │               186 │        3800 │ female │  2007 │
│ Adelie  │ Torgersen │           40.3 │          18.0 │               195 │        3250 │ female │  2007 │
│ Adelie  │ Torgersen │           39.3 │          20.6 │               190 │        3650 │ male   │  2007 │
│ Adelie  │ Torgersen │           38.9 │          17.8 │               181 │        3625 │ female │  2007 │
│ Adelie  │ Torgersen │           39.2 │          19.6 │               195 │        4675 │ male   │  2007 │
│ Adelie  │ Torgersen │           42.0 │          20.2 │               190 │        4250 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           37.8 │          17.1 │               186 │        3300 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           37.8 │          17.3 │               180 │        3700 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           41.1 │          17.6 │               182 │        3200 │ female │  2007 │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Of course, the filtering expression can be applied inline as well.

In [33]:

filtered = penguins[penguins.bill_length_mm > 37.0]
filtered

filtered = penguins[penguins.bill_length_mm > 37.0] filtered

Out[33]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.5 │          17.4 │               186 │        3800 │ female │  2007 │
│ Adelie  │ Torgersen │           40.3 │          18.0 │               195 │        3250 │ female │  2007 │
│ Adelie  │ Torgersen │           39.3 │          20.6 │               190 │        3650 │ male   │  2007 │
│ Adelie  │ Torgersen │           38.9 │          17.8 │               181 │        3625 │ female │  2007 │
│ Adelie  │ Torgersen │           39.2 │          19.6 │               195 │        4675 │ male   │  2007 │
│ Adelie  │ Torgersen │           42.0 │          20.2 │               190 │        4250 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           37.8 │          17.1 │               186 │        3300 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           37.8 │          17.3 │               180 │        3700 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           41.1 │          17.6 │               182 │        3200 │ female │  2007 │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Multiple filtering expressions can be combined into a single expression or chained onto existing table expressions.

In [34]:

filtered = penguins[(penguins.bill_length_mm > 37.0) & (penguins.bill_depth_mm > 18.0)]
filtered

filtered = penguins[(penguins.bill_length_mm > 37.0) & (penguins.bill_depth_mm > 18.0)] filtered

Out[34]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.3 │          20.6 │               190 │        3650 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.2 │          19.6 │               195 │        4675 │ male   │  2007 │
│ Adelie  │ Torgersen │           42.0 │          20.2 │               190 │        4250 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           38.6 │          21.2 │               191 │        3800 │ male   │  2007 │
│ Adelie  │ Torgersen │           38.7 │          19.0 │               195 │        3450 │ female │  2007 │
│ Adelie  │ Torgersen │           42.5 │          20.7 │               197 │        4500 │ male   │  2007 │
│ Adelie  │ Torgersen │           46.0 │          21.5 │               194 │        4200 │ male   │  2007 │
│ Adelie  │ Biscoe    │           37.8 │          18.3 │               174 │        3400 │ female │  2007 │
│ Adelie  │ Biscoe    │           37.7 │          18.7 │               180 │        3600 │ male   │  2007 │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

The code above will return the same rows as the code below.

In [35]:

filtered = penguins[penguins.bill_length_mm > 37.0][penguins.bill_depth_mm > 18.0]
filtered

filtered = penguins[penguins.bill_length_mm > 37.0][penguins.bill_depth_mm > 18.0] filtered

Out[35]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.3 │          20.6 │               190 │        3650 │ male   │  2007 │
│ Adelie  │ Torgersen │           39.2 │          19.6 │               195 │        4675 │ male   │  2007 │
│ Adelie  │ Torgersen │           42.0 │          20.2 │               190 │        4250 │ NULL   │  2007 │
│ Adelie  │ Torgersen │           38.6 │          21.2 │               191 │        3800 │ male   │  2007 │
│ Adelie  │ Torgersen │           38.7 │          19.0 │               195 │        3450 │ female │  2007 │
│ Adelie  │ Torgersen │           42.5 │          20.7 │               197 │        4500 │ male   │  2007 │
│ Adelie  │ Torgersen │           46.0 │          21.5 │               194 │        4200 │ male   │  2007 │
│ Adelie  │ Biscoe    │           37.8 │          18.3 │               174 │        3400 │ female │  2007 │
│ Adelie  │ Biscoe    │           37.7 │          18.7 │               180 │        3600 │ male   │  2007 │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Aggregation has not been discussed yet, but aggregate values can be used in expressions to return things such as all of the rows in a data set where the value in a column is greater than the mean.

In [36]:

filtered = penguins[penguins.bill_length_mm > penguins.bill_length_mm.mean()]
filtered

filtered = penguins[penguins.bill_length_mm > penguins.bill_length_mm.mean()] filtered

Out[36]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Torgersen │           46.0 │          21.5 │               194 │        4200 │ male   │  2007 │
│ Adelie  │ Dream     │           44.1 │          19.7 │               196 │        4400 │ male   │  2007 │
│ Adelie  │ Torgersen │           45.8 │          18.9 │               197 │        4150 │ male   │  2008 │
│ Adelie  │ Biscoe    │           45.6 │          20.3 │               191 │        4600 │ male   │  2009 │
│ Adelie  │ Torgersen │           44.1 │          18.0 │               210 │        4000 │ male   │  2009 │
│ Gentoo  │ Biscoe    │           46.1 │          13.2 │               211 │        4500 │ female │  2007 │
│ Gentoo  │ Biscoe    │           50.0 │          16.3 │               230 │        5700 │ male   │  2007 │
│ Gentoo  │ Biscoe    │           48.7 │          14.1 │               210 │        4450 │ female │  2007 │
│ Gentoo  │ Biscoe    │           50.0 │          15.2 │               218 │        5700 │ male   │  2007 │
│ Gentoo  │ Biscoe    │           47.6 │          14.5 │               215 │        5400 │ male   │  2007 │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Modifying rows¶

Sometimes you want to modify the values in a column based on some condition. In pandas, you would do something like df.loc[condition] = new_value. In Ibis though, remember that all expressions are immutable, so you need to create a new table expression with the modified values. You do this using the ifelse method on boolean columns:

In [37]:

long_billed_penguins = penguins.bill_length_mm > 37.0
species_modified = long_billed_penguins.ifelse('wide', penguins.species)
penguins.mutate(species_modified=species_modified)

long_billed_penguins = penguins.bill_length_mm > 37.0 species_modified = long_billed_penguins.ifelse('wide', penguins.species) penguins.mutate(species_modified=species_modified)

Out[37]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┳━━━━━━━━━━━━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃ species_modified ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━╇━━━━━━━━━━━━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │ string           │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┼──────────────────┤
│ Adelie  │ Torgersen │           39.1 │          18.7 │               181 │        3750 │ male   │  2007 │ wide             │
│ Adelie  │ Torgersen │           39.5 │          17.4 │               186 │        3800 │ female │  2007 │ wide             │
│ Adelie  │ Torgersen │           40.3 │          18.0 │               195 │        3250 │ female │  2007 │ wide             │
│ Adelie  │ Torgersen │            nan │           nan │              NULL │        NULL │ NULL   │  2007 │ Adelie           │
│ Adelie  │ Torgersen │           36.7 │          19.3 │               193 │        3450 │ female │  2007 │ Adelie           │
│ Adelie  │ Torgersen │           39.3 │          20.6 │               190 │        3650 │ male   │  2007 │ wide             │
│ Adelie  │ Torgersen │           38.9 │          17.8 │               181 │        3625 │ female │  2007 │ wide             │
│ Adelie  │ Torgersen │           39.2 │          19.6 │               195 │        4675 │ male   │  2007 │ wide             │
│ Adelie  │ Torgersen │           34.1 │          18.1 │               193 │        3475 │ NULL   │  2007 │ Adelie           │
│ Adelie  │ Torgersen │           42.0 │          20.2 │               190 │        4250 │ NULL   │  2007 │ wide             │
│ …       │ …         │              … │             … │                 … │           … │ …      │     … │ …                │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┴──────────────────┘

Sorting rows¶

Sorting rows in Ibis uses a somewhat different API than in pandas. In pandas, you would use the sort_values method to order rows by values in specified columns. Ibis uses a method called order_by. To specify ascending or descending orders, pandas uses an ascending= argument to sort_values that indicates the order for each sorting column. Ibis allows you to tag the column name in the order_by list as ascending or descending by wrapping it with ibis.asc or ibis.desc.

First, let's ask Ibis for a pandas DataFrame version of the penguin data:

In [38]:

df = penguins.to_pandas()

df = penguins.to_pandas()

Here is an example of sorting a DataFrame using two sort keys. One key is sorting in ascending order and the other is in descending order.

In [39]:

df.sort_values(
    ['bill_length_mm', 'bill_depth_mm'], ascending=[True, False], na_position="first"
).head(5)

df.sort_values( ['bill_length_mm', 'bill_depth_mm'], ascending=[True, False], na_position="first" ).head(5)

Out[39]:

	species	island	bill_length_mm	bill_depth_mm	flipper_length_mm	body_mass_g	sex	year
3	Adelie	Torgersen	NaN	NaN	NaN	NaN	None	2007
271	Gentoo	Biscoe	NaN	NaN	NaN	NaN	None	2009
142	Adelie	Dream	32.1	15.5	188.0	3050.0	female	2009
98	Adelie	Dream	33.1	16.1	178.0	2900.0	female	2008
70	Adelie	Torgersen	33.5	19.0	190.0	3600.0	female	2008

The same operation in Ibis would look like the following. Note that the index values of the resulting DataFrame start from zero and count up, whereas in the example above, they retain their original index value. This is simply due to the fact that rows in tables don't necessarily have a stable index in database backends, so the index is just generated on the result.

In [40]:

sorted = penguins.order_by(['bill_length_mm', ibis.desc('bill_depth_mm')]).head(5)
sorted

sorted = penguins.order_by(['bill_length_mm', ibis.desc('bill_depth_mm')]).head(5) sorted

Out[40]:

┏━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━┳━━━━━━━┓
┃ species ┃ island    ┃ bill_length_mm ┃ bill_depth_mm ┃ flipper_length_mm ┃ body_mass_g ┃ sex    ┃ year  ┃
┡━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━╇━━━━━━━┩
│ string  │ string    │ float64        │ float64       │ int64             │ int64       │ string │ int64 │
├─────────┼───────────┼────────────────┼───────────────┼───────────────────┼─────────────┼────────┼───────┤
│ Adelie  │ Dream     │           32.1 │          15.5 │               188 │        3050 │ female │  2009 │
│ Adelie  │ Dream     │           33.1 │          16.1 │               178 │        2900 │ female │  2008 │
│ Adelie  │ Torgersen │           33.5 │          19.0 │               190 │        3600 │ female │  2008 │
│ Adelie  │ Dream     │           34.0 │          17.1 │               185 │        3400 │ female │  2008 │
│ Adelie  │ Torgersen │           34.1 │          18.1 │               193 │        3475 │ NULL   │  2007 │
└─────────┴───────────┴────────────────┴───────────────┴───────────────────┴─────────────┴────────┴───────┘

Aggregation¶

Aggregation in pandas is typically done by computing columns based on an aggregate function.

In [41]:

stats = [df.bill_depth_mm.sum(), df.bill_length_mm.mean()]
pd.DataFrame([stats], columns=['total_bill_depth', 'avg.bill_length'])

stats = [df.bill_depth_mm.sum(), df.bill_length_mm.mean()] pd.DataFrame([stats], columns=['total_bill_depth', 'avg.bill_length'])

Out[41]:

	total_bill_depth	avg.bill_length
0	5865.7	43.92193

In Ibis, you construct aggregate expressions then apply them to the table using the aggregate method.

In [42]:

stats = [
    penguins.bill_depth_mm.sum().name('total_bill_width'),
    penguins.bill_length_mm.mean().name('avg_bill_length'),
]
agged = penguins.aggregate(stats)
agged

stats = [ penguins.bill_depth_mm.sum().name('total_bill_width'), penguins.bill_length_mm.mean().name('avg_bill_length'), ] agged = penguins.aggregate(stats) agged

Out[42]:

┏━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ total_bill_width ┃ avg_bill_length ┃
┡━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ float64          │ float64         │
├──────────────────┼─────────────────┤
│           5865.7 │        43.92193 │
└──────────────────┴─────────────────┘

You can also combine both operations into one and pass the aggregate expressions using keyword parameters.

In [43]:

agged = penguins.aggregate(
    total_bill_depth=penguins.bill_depth_mm.sum(),
    avg_bill_length=penguins.bill_length_mm.mean(),
)
agged

agged = penguins.aggregate( total_bill_depth=penguins.bill_depth_mm.sum(), avg_bill_length=penguins.bill_length_mm.mean(), ) agged

Out[43]:

┏━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ total_bill_depth ┃ avg_bill_length ┃
┡━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ float64          │ float64         │
├──────────────────┼─────────────────┤
│           5865.7 │        43.92193 │
└──────────────────┴─────────────────┘

Group by¶

Aggregations can also be done across groupings using the by= parameter.

In [44]:

agged = penguins.aggregate(
    by='species',
    total_bill_depth=penguins.bill_depth_mm.sum(),
    avg_bill_length=penguins.bill_length_mm.mean(),
)
agged

agged = penguins.aggregate( by='species', total_bill_depth=penguins.bill_depth_mm.sum(), avg_bill_length=penguins.bill_length_mm.mean(), ) agged

Out[44]:

┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ species   ┃ total_bill_depth ┃ avg_bill_length ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ string    │ float64          │ float64         │
├───────────┼──────────────────┼─────────────────┤
│ Adelie    │           2770.3 │       38.791391 │
│ Gentoo    │           1842.8 │       47.504878 │
│ Chinstrap │           1252.6 │       48.833824 │
└───────────┴──────────────────┴─────────────────┘

Alternatively, by groups can be computed using a grouped table.

In [45]:

agged = penguins.group_by('species').aggregate(
    total_bill_depth=penguins.bill_depth_mm.sum(),
    avg_bill_length=penguins.bill_length_mm.mean(),
)
agged

agged = penguins.group_by('species').aggregate( total_bill_depth=penguins.bill_depth_mm.sum(), avg_bill_length=penguins.bill_length_mm.mean(), ) agged

Out[45]:

┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━┓
┃ species   ┃ total_bill_depth ┃ avg_bill_length ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━┩
│ string    │ float64          │ float64         │
├───────────┼──────────────────┼─────────────────┤
│ Adelie    │           2770.3 │       38.791391 │
│ Gentoo    │           1842.8 │       47.504878 │
│ Chinstrap │           1252.6 │       48.833824 │
└───────────┴──────────────────┴─────────────────┘

Dropping rows with NULLs¶

Both pandas and Ibis allow you to drop rows from a table based on whether a set of columns contains a NULL value. This method is called dropna in both packages. The common set of parameters in the two are subset= and how=. The subset= parameter indicates which columns to inspect for NULL values. The how= parameter specifies whether 'any' or 'all' of the specified columns must be NULL in order for the row to be dropped.

In [46]:

no_null_peng = penguins.dropna(['bill_depth_mm', 'bill_length_mm'], how='any')

no_null_peng = penguins.dropna(['bill_depth_mm', 'bill_length_mm'], how='any')

Filling NULL values¶

Both pandas and Ibis allow you to fill NULL values in a table. In Ibis, the replacement value can only be a scalar value of a dictionary of values. If it is a dictionary, the keys of the dictionary specify the column name for the value to apply to.

In [47]:

no_null_peng = penguins.fillna(dict(bill_depth_mm=0, bill_length_mm=0))

no_null_peng = penguins.fillna(dict(bill_depth_mm=0, bill_length_mm=0))

Common column expressions¶

See the full API documentation for all of the available value methods and tools for creating value expressions. We mention a few common ones here as they relate to common SQL queries.

Type casts¶

Type casting in pandas is done using the astype method on columns.

In [48]:

df.bill_depth_mm.astype(str)

df.bill_depth_mm.astype(str)

Out[48]:

0      18.7
1      17.4
2      18.0
3       nan
4      19.3
       ... 
339    19.8
340    18.1
341    18.2
342    19.0
343    18.7
Name: bill_depth_mm, Length: 344, dtype: object

In Ibis, you cast the column type using the cast method.

In [49]:

penguins.bill_depth_mm.cast('int')

penguins.bill_depth_mm.cast('int')

Out[49]:

┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Cast(bill_depth_mm, int64) ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ int64                      │
├────────────────────────────┤
│                         19 │
│                         17 │
│                         18 │
│                       NULL │
│                         19 │
│                         21 │
│                         18 │
│                         20 │
│                         18 │
│                         20 │
│                          … │
└────────────────────────────┘

Casted columns can be assigned back to the table using the mutate method described earlier.

In [50]:

casted = penguins.mutate(
    bill_depth_mm=penguins.bill_depth_mm.cast('int'),
    bill_length_mm=penguins.bill_length_mm.cast('int'),
)
casted.schema()

casted = penguins.mutate( bill_depth_mm=penguins.bill_depth_mm.cast('int'), bill_length_mm=penguins.bill_length_mm.cast('int'), ) casted.schema()

Out[50]:

ibis.Schema {
  species            string
  island             string
  bill_length_mm     int64
  bill_depth_mm      int64
  flipper_length_mm  int64
  body_mass_g        int64
  sex                string
  year               int64
}

Replacing NULLs¶

Both pandas and Ibis have fillna methods which allow you to specify a replacement value for NULL values.

In [51]:

bill_length_no_nulls = penguins.bill_length_mm.fillna(0)

bill_length_no_nulls = penguins.bill_length_mm.fillna(0)

Set membership¶

pandas set membership uses the in and not in operators such as 'a' in df.species. Ibis uses isin and notin methods. In addition to testing membership in a set, these methods allow you to specify an else case to assign a value when the value isn't in the set.

In [52]:

penguins.species.value_counts()

penguins.species.value_counts()

Out[52]:

┏━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ species   ┃ species_count ┃
┡━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ string    │ int64         │
├───────────┼───────────────┤
│ Adelie    │           152 │
│ Gentoo    │           124 │
│ Chinstrap │            68 │
└───────────┴───────────────┘

In [53]:

refined = penguins.species.isin(['Adelie', 'Chinstrap'])
refined.value_counts()

refined = penguins.species.isin(['Adelie', 'Chinstrap']) refined.value_counts()

Out[53]:

┏━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Contains(species) ┃ Contains(species)_count ┃
┡━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ boolean           │ int64                   │
├───────────────────┼─────────────────────────┤
│ True              │                     220 │
│ False             │                     124 │
└───────────────────┴─────────────────────────┘

Merging tables¶

While pandas uses the merge method to combine data from multiple DataFrames, Ibis uses the join method. They both have similar capabilities.

The biggest difference between Ibis' join method and pandas' merge method is that pandas only accepts column names or index levels to join on, whereas Ibis can merge on expressions.

Here are some examples of merging using pandas.

In [54]:

df_left = pd.DataFrame(
    [
        ['a', 1, 2],
        ['b', 3, 4],
        ['c', 4, 6],
    ],
    columns=['name', 'x', 'y'],
)

df_right = pd.DataFrame(
    [
        ['a', 100, 200],
        ['m', 300, 400],
        ['n', 400, 600],
    ],
    columns=['name', 'x_100', 'y_100'],
)

df_left = pd.DataFrame( [ ['a', 1, 2], ['b', 3, 4], ['c', 4, 6], ], columns=['name', 'x', 'y'], ) df_right = pd.DataFrame( [ ['a', 100, 200], ['m', 300, 400], ['n', 400, 600], ], columns=['name', 'x_100', 'y_100'], )

In [55]:

df_left.merge(df_right, on='name')

df_left.merge(df_right, on='name')

Out[55]:

	name	x	y	x_100	y_100
0	a	1	2	100	200

In [56]:

df_left.merge(df_right, on='name', how='outer')

df_left.merge(df_right, on='name', how='outer')

Out[56]:

	name	x	y	x_100	y_100
0	a	1.0	2.0	100.0	200.0
1	b	3.0	4.0	NaN	NaN
2	c	4.0	6.0	NaN	NaN
3	m	NaN	NaN	300.0	400.0
4	n	NaN	NaN	400.0	600.0

We can now convert DataFrames to Ibis tables to do joins.

In [57]:

t_left = ibis.memtable(df_left, name="t_left")
t_right = ibis.memtable(df_right, name='t_right')

t_left = ibis.memtable(df_left, name="t_left") t_right = ibis.memtable(df_right, name='t_right')

In [58]:

t_left.join(t_right, t_left.name == t_right.name)

t_left.join(t_right, t_left.name == t_right.name)

Out[58]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ name   ┃ x     ┃ y     ┃ x_100 ┃ y_100 ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │ int64 │ int64 │
├────────┼───────┼───────┼───────┼───────┤
│ a      │     1 │     2 │   100 │   200 │
└────────┴───────┴───────┴───────┴───────┘

Below is an outer join where missing values are filled with NaN.

In [59]:

t_left.join(t_right, t_left.name == t_right.name, how='outer')

t_left.join(t_right, t_left.name == t_right.name, how='outer')

Out[59]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ name   ┃ x     ┃ y     ┃ name_right ┃ x_100 ┃ y_100 ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │ string     │ int64 │ int64 │
├────────┼───────┼───────┼────────────┼───────┼───────┤
│ a      │     1 │     2 │ a          │   100 │   200 │
│ b      │     3 │     4 │ NULL       │  NULL │  NULL │
│ c      │     4 │     6 │ NULL       │  NULL │  NULL │
│ NULL   │  NULL │  NULL │ m          │   300 │   400 │
│ NULL   │  NULL │  NULL │ n          │   400 │   600 │
└────────┴───────┴───────┴────────────┴───────┴───────┘

Concatenating tables¶

Concatenating DataFrames in pandas is done with the concat top-level function. It takes multiple DataFrames and concatenates the rows of one DataFrame to the next. If the columns are mis-matched, it extends the list of columns to include the full set of columns and inserts NaNs and Nones into the missing values.

Concatenating tables in Ibis can only be done on tables with matching schemas. The concatenation is done using the top-level union function or the union method on a table.

We'll demonstrate a pandas concat first.

In [60]:

df_1 = pd.DataFrame(
    [
        ['a', 1, 2],
        ['b', 3, 4],
        ['c', 4, 6],
    ],
    columns=['name', 'x', 'y'],
)

df_2 = pd.DataFrame(
    [
        ['a', 100, 200],
        ['m', 300, 400],
        ['n', 400, 600],
    ],
    columns=['name', 'x', 'y'],
)

df_1 = pd.DataFrame( [ ['a', 1, 2], ['b', 3, 4], ['c', 4, 6], ], columns=['name', 'x', 'y'], ) df_2 = pd.DataFrame( [ ['a', 100, 200], ['m', 300, 400], ['n', 400, 600], ], columns=['name', 'x', 'y'], )

In [61]:

pd.concat([df_1, df_2])

pd.concat([df_1, df_2])

Out[61]:

	name	x	y
0	a	1	2
1	b	3	4
2	c	4	6
0	a	100	200
1	m	300	400
2	n	400	600

Now we can convert the DataFrames to Ibis tables and combine the tables using a union.

In [62]:

t_1 = ibis.memtable(df_1, name='t_1')
t_2 = ibis.memtable(df_2, name='t_2')

t_1 = ibis.memtable(df_1, name='t_1') t_2 = ibis.memtable(df_2, name='t_2')

In [63]:

unioned = ibis.union(t_1, t_2)
unioned

unioned = ibis.union(t_1, t_2) unioned

Out[63]:

┏━━━━━━━━┳━━━━━━━┳━━━━━━━┓
┃ name   ┃ x     ┃ y     ┃
┡━━━━━━━━╇━━━━━━━╇━━━━━━━┩
│ string │ int64 │ int64 │
├────────┼───────┼───────┤
│ a      │     1 │     2 │
│ b      │     3 │     4 │
│ c      │     4 │     6 │
│ a      │   100 │   200 │
│ m      │   300 │   400 │
│ n      │   400 │   600 │
└────────┴───────┴───────┘

---

Last update: May 8, 2023