Skip to content

It is a plugin to pyexcel and provides the capability to present and write data in text formats using tabulate

License

Notifications You must be signed in to change notification settings

pyexcel-renderers/pyexcel-text

Repository files navigation

pyexcel-text - Let you focus on data, instead of text formats

https://raw.githubusercontent.com/pyexcel/pyexcel.github.io/master/images/patreon.png https://api.travis-ci.org/pyexcel/pyexcel-text.svg?branch=master

It is a plugin to pyexcel and extends its capbility to present and write data in text fromats mainly through tabulate:

  • "plain"
  • "simple"
  • "grid"
  • "pipe"
  • "orgtbl"
  • "rst"
  • "mediawiki"
  • "latex"
  • "latex_booktabs"
  • "json"
  • "html"

Since v0.2.7, json and ndjson input are also supported.

Usage

What's new?

>>> import pyexcel as pe
>>> sheet = pe.Sheet()
>>> sheet.json = '[[1,2],[2,3]]'
>>> sheet
pyexcel sheet:
+---+---+
| 1 | 2 |
+---+---+
| 2 | 3 |
+---+---+
>>> highspeedrail = pe.Sheet()
>>> highspeedrail.json = """
... [{"year": 1903, "country": "Germany", "speed": "206.7km/h"},
... {"year": 1964, "country": "Japan", "speed": "210km/h"},
... {"year": 2008, "country": "China", "speed": "350km/h"}]
... """
>>> highspeedrail.name = 'High Speed Train Speed Break Through (Source: Wikipedia)'
>>> highspeedrail
High Speed Train Speed Break Through (Source: Wikipedia):
+---------+-----------+------+
| country | speed     | year |
+---------+-----------+------+
| Germany | 206.7km/h | 1903 |
+---------+-----------+------+
| Japan   | 210km/h   | 1964 |
+---------+-----------+------+
| China   | 350km/h   | 2008 |
+---------+-----------+------+
>>> henley_on_thames_facts = pe.Sheet()
>>> henley_on_thames_facts.json = """
... {"area": "5.58 square meters",
... "population": "11,619",
... "civial parish": "Henley-on-Thames",
... "latitude": "51.536",
... "longitude": "-0.898"
... }"""
>>> henley_on_thames_facts
pyexcel sheet:
+--------------------+------------------+----------+-----------+------------+
| area               | civial parish    | latitude | longitude | population |
+--------------------+------------------+----------+-----------+------------+
| 5.58 square meters | Henley-on-Thames | 51.536   | -0.898    | 11,619     |
+--------------------+------------------+----------+-----------+------------+
>>> ccs_insight = pe.Sheet()
>>> ccs_insight.name = "Worldwide Mobile Phone Shipments (Billions), 2017-2021"
>>> ccs_insight.json = """
... {"year": ["2017", "2018", "2019", "2020", "2021"],
... "smart phones": [1.53, 1.64, 1.74, 1.82, 1.90],
... "feature phones": [0.46, 0.38, 0.30, 0.23, 0.17]}"""
>>> ccs_insight
pyexcel sheet:
+----------------+--------------+------+
| feature phones | smart phones | year |
+----------------+--------------+------+
| 0.46           | 1.53         | 2017 |
+----------------+--------------+------+
| 0.38           | 1.64         | 2018 |
+----------------+--------------+------+
| 0.3            | 1.74         | 2019 |
+----------------+--------------+------+
| 0.23           | 1.82         | 2020 |
+----------------+--------------+------+
| 0.17           | 1.9          | 2021 |
+----------------+--------------+------+

Here is a variant of json:

>>> highspeedrail2 = pe.Sheet()
>>> highspeedrail2.ndjson = """
... {"year": 1903, "country": "Germany", "speed": "206.7km/h"}
... {"year": 1964, "country": "Japan", "speed": "210km/h"}
... {"year": 2008, "country": "China", "speed": "350km/h"}
... """.strip()
>>> highspeedrail2.name = 'High Speed Train Speed Break Through (Source: Wikipedia)'
>>> highspeedrail2
High Speed Train Speed Break Through (Source: Wikipedia):
+---------+-----------+------+
| country | speed     | year |
+---------+-----------+------+
| Germany | 206.7km/h | 1903 |
+---------+-----------+------+
| Japan   | 210km/h   | 1964 |
+---------+-----------+------+
| China   | 350km/h   | 2008 |
+---------+-----------+------+
>>> henley_on_thames_facts2 = pe.Sheet()
>>> henley_on_thames_facts2.ndjson = """
... {"area": "5.58 square meters"}
... {"population": "11,619"}
... {"civial parish": "Henley-on-Thames"}
... {"latitude": "51.536"}
... {"longitude": "-0.898"}
... """.strip()
>>> henley_on_thames_facts2
pyexcel sheet:
+---------------+--------------------+
| area          | 5.58 square meters |
+---------------+--------------------+
| population    | 11,619             |
+---------------+--------------------+
| civial parish | Henley-on-Thames   |
+---------------+--------------------+
| latitude      | 51.536             |
+---------------+--------------------+
| longitude     | -0.898             |
+---------------+--------------------+
>>> ccs_insight2 = pe.Sheet()
>>> ccs_insight2.name = "Worldwide Mobile Phone Shipments (Billions), 2017-2021"
>>> ccs_insight2.ndjson = """
... {"year": ["2017", "2018", "2019", "2020", "2021"]}
... {"smart phones": [1.53, 1.64, 1.74, 1.82, 1.90]}
... {"feature phones": [0.46, 0.38, 0.30, 0.23, 0.17]}
... """.strip()
>>> ccs_insight2
pyexcel sheet:
+----------------+------+------+------+------+------+
| year           | 2017 | 2018 | 2019 | 2020 | 2021 |
+----------------+------+------+------+------+------+
| smart phones   | 1.53 | 1.64 | 1.74 | 1.82 | 1.9  |
+----------------+------+------+------+------+------+
| feature phones | 0.46 | 0.38 | 0.3  | 0.23 | 0.17 |
+----------------+------+------+------+------+------+

Simple

>>> import pyexcel as pe
>>> content = [
...     ["Column 1", "Column 2", "Column 3"],
...     [1, 2, 3],
...     [4, 5, 6],
...     [7, 8, 9]
... ]
>>> sheet = pe.Sheet(content)
>>> print(sheet.simple)
pyexcel sheet:
--------  --------  --------
Column 1  Column 2  Column 3
1         2         3
4         5         6
7         8         9
--------  --------  --------
>>> sheet.name_columns_by_row(0)
>>> print(sheet.simple)
pyexcel sheet:
  Column 1    Column 2    Column 3
----------  ----------  ----------
         1           2           3
         4           5           6
         7           8           9

Grid

>>> print(sheet.grid)
pyexcel sheet:
+------------+------------+------------+
|   Column 1 |   Column 2 |   Column 3 |
+============+============+============+
|          1 |          2 |          3 |
+------------+------------+------------+
|          4 |          5 |          6 |
+------------+------------+------------+
|          7 |          8 |          9 |
+------------+------------+------------+

Mediawiki

>>> multiple_sheets = {
...      'Sheet 1':
...          [
...              [1.0, 2.0, 3.0],
...              [4.0, 5.0, 6.0],
...              [7.0, 8.0, 9.0]
...          ],
...      'Sheet 2':
...          [
...              ['X', 'Y', 'Z'],
...              [1.0, 2.0, 3.0],
...              [4.0, 5.0, 6.0]
...          ],
...      'Sheet 3':
...          [
...              ['O', 'P', 'Q'],
...              [3.0, 2.0, 1.0],
...              [4.0, 3.0, 2.0]
...          ]
...  }
>>> book = pe.Book(multiple_sheets)
>>> book.save_as("myfile.mediawiki")
>>> myfile = open("myfile.mediawiki")
>>> print(myfile.read())
Sheet 1:
{| class="wikitable" style="text-align: left;"
|+ <!-- caption -->
|-
| align="right"| 1 || align="right"| 2 || align="right"| 3
|-
| align="right"| 4 || align="right"| 5 || align="right"| 6
|-
| align="right"| 7 || align="right"| 8 || align="right"| 9
|}
Sheet 2:
{| class="wikitable" style="text-align: left;"
|+ <!-- caption -->
|-
| X   || Y   || Z
|-
| 1.0 || 2.0 || 3.0
|-
| 4.0 || 5.0 || 6.0
|}
Sheet 3:
{| class="wikitable" style="text-align: left;"
|+ <!-- caption -->
|-
| O   || P   || Q
|-
| 3.0 || 2.0 || 1.0
|-
| 4.0 || 3.0 || 2.0
|}
>>> myfile.close()

Html

>>> book.save_as("myfile.html")
>>> myfile = open("myfile.html")
>>> print(myfile.read()) # doctest: +SKIP
Sheet 1:
<table>
<tr><td style="text-align: right;">1</td><td style="text-align: right;">2</td><td style="text-align: right;">3</td></tr>
<tr><td style="text-align: right;">4</td><td style="text-align: right;">5</td><td style="text-align: right;">6</td></tr>
<tr><td style="text-align: right;">7</td><td style="text-align: right;">8</td><td style="text-align: right;">9</td></tr>
</table>
Sheet 2:
<table>
<tr><td>X  </td><td>Y  </td><td>Z  </td></tr>
<tr><td>1.0</td><td>2.0</td><td>3.0</td></tr>
<tr><td>4.0</td><td>5.0</td><td>6.0</td></tr>
</table>
Sheet 3:
<table>
<tr><td>O  </td><td>P  </td><td>Q  </td></tr>
<tr><td>3.0</td><td>2.0</td><td>1.0</td></tr>
<tr><td>4.0</td><td>3.0</td><td>2.0</td></tr>
</table>

Please note tabulate 0.7.7 gives an extra tbody tag around tr tag.

.. testcode::
   :hide:

    >>> myfile.close()
    >>> import os
    >>> os.unlink("myfile.mediawiki")
    >>> os.unlink("myfile.html")


Dependencies

  • tabulate

About

It is a plugin to pyexcel and provides the capability to present and write data in text formats using tabulate

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages