Change nltk.download() path directory from default ~/ntlk_data

Question

I was trying to download/update python nltk packages on a computing server and it returned this [Errno 122] Disk quota exceeded: error.

Specifically:

[nltk_data] Downloading package stop words to /home/sh2264/nltk_data...
[nltk_data] Error downloading u'stopwords' from
[nltk_data] <https://raw.githubusercontent.com/nltk/nltk_data/gh-
[nltk_data] pages/packages/corpora/stopwords.zip>: [Errno 122]
[nltk_data] Disk quota exceeded:
[nltk_data] u'/home/sh2264/nltk_data/corpora/stopwords.zip
False

How could I change the entire path for nltk packages, and what other changes should I make to ensure errorless loading of nltk?

smci · Accepted Answer · 2017-11-20T03:44:26.023

This can be configured both by command-line (nltk.download(..., download_dir=) or by GUI. Bizarrely nltk seems to totally ignore its own environment variable NLTK_DATA and default its download directories to a standard set of five paths, regardless whether NLTK_DATA is defined and where it points, and regardless whether nltk's five default dirs even exist on the machine or architecture(!). Some of that is documented in Installing NLTK Data, although it's incomplete and kinda buried; reproduced below with much clearer formatting:

Command line installation

The downloader will search for an existing nltk_data directory to install NLTK data. If one does not exist it will attempt to create one in a central location (when using an administrator account) or otherwise in the user’s filespace. If necessary, run the download command from an administrator account, or using sudo. The recommended system location is:

C:\nltk_data (Windows) ;

/usr/local/share/nltk_data (Mac) and

/usr/share/nltk_data (Unix).

You can use the -d flag to specify a different location (but if you do this, be sure to set the NLTK_DATA environment variable accordingly).

Run the command python -m nltk.downloader all

To ensure central installation, run the command: sudo python -m nltk.downloader -d /usr/local/share/nltk_data all

But really they should say: sudo python -m nltk.downloader -d $NLTK_DATA all

Now as to what recommended path NLTK_DATA should use, nltk doesn't really give any proper guidance, but it should be a generic standalone path not under any install tree (so not under <python-install-directory>/lib/site-packages) or any user dir. Hence, /usr/local/share, /opt/share or similar. On MacOS 10.7+, /usr and thus /usr/local/ these days are hidden by default, so /opt/share may well be a better choice. Or do chflags nohidden /usr/local/share.

This is not the behavior I see.. as root in a docker container it downloads into /root/nltk_data. — user239558, Feb 07 '18 at 08:06
If we download only stopwords for the specific directory in Linux `sudo python -m nltk.downloader -d /usr/local/share/nltk_data stopwords` — Sashini Hettiarachchi, Jan 31 '19 at 18:17
@HansikaHettiarachchi you can specify more than one download. `sudo python -m nltk.downloader -d /usr/local/share/nltk_data stopwords wordnet punkt` — nbeuchat, Feb 22 '19 at 20:32
@user239558 et al, if you found a docbug, please [report it to nltk](https://github.com/nltk/nltk/wiki/FAQ) — smci, Feb 22 '19 at 21:41
Just to avoid confusion: you need to specify the path including the `nltk_data` part, so if you want to install in `/usr/share` then the command (within the Python shell) is `nltk.download('...',download_dir='/usr/share/nltk_data')`. — Marijn, Sep 26 '21 at 10:08

score 27 · Answer 2 · answered Jul 06 '17 at 13:11

27

According to the documentation:

By default, packages are installed in either a system-wide directory (if Python has sufficient access to write to it); or in the current user’s home directory. However, the download_dir argument may be used to specify a different installation target, if desired.

To specify the download directory, use for example:

nltk.download('treebank', download_dir='/mnt/data/treebank')

answered Jul 06 '17 at 13:11

Ortomala Lokni

56,620
24
188
240

11

To retrieve some nltk's downloaded packages you may need to also include the `download_dir` to nltk's data path -> `nltk.data.path.append('/mnt/data/treebank')` – Makan Sep 10 '20 at 17:56

score 3 · Answer 3 · answered Jan 25 '19 at 07:45

3

You may also use nltk.download_shell() and follow the interactive steps as shown below.

Also use nltk.data.path.append('/your/new/data/directory/path') to instruct nltk to to load data from new data path.

answered Jan 25 '19 at 07:45

webstruck

611
7
9

score 0 · Answer 4 · answered Jul 23 '18 at 17:47

NLTK GUI can be started from PyCharm Community Edition Python console too. Just issue 2 commands:

1) import nltk

2) nltk.download_gui()

but nltk GUI will not work if you are behind a proxy server for that at the console you must first set proxy setting

SET HTTP_PROXY=proxy.mycompany.com:8080

and then it will work.

score 0 · Answer 5 · answered Jul 23 '22 at 06:12

This would be the solution purely from shell commands (no python code).

First use this command line to download to a custom directory:

python -m nltk.downloader -d /my/path/nltk_data all

And then at runtime for nltk to find the custom directory, set the environment variable:

export NLTK_DATA=/my/path/nltk_data

score 0 · Answer 6 · answered Sep 15 '22 at 13:11

tl;dr

If the environment variable NLTK_DATA is set and the directory exists, it is used as default download directory.

Explanation

In the data module the environment variable NLTK_DATA is used as first entry when filling the data search path.

If download_dir is not specified as parameter when calling nltk.download() the method default_download_dir() determines the download directory.

Example: Create new default data directory

One would like to use /usr/local/share/nltk_data as default data directory.

Create the data directory and add NLTK_DATA to your shell profile.

$ mkdir /usr/local/share/nltk_data
$ echo "export NLTK_DATA=/usr/local/share/nltk_data" >> ~/.bashrc
$ source ~/.bashrc
$ echo $NLTK_DATA
/usr/local/share/nltk_data

Now nltk uses /usr/local/share/nltk_data as defined in NLTK_DATA.

$ python
Python 3.10.6 (main, Sep  5 2022, 11:08:58) [Clang 11.0.0 (clang-1100.0.33.17)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import nltk
>>> nltk.download('gutenberg')
[nltk_data] Downloading package gutenberg to
[nltk_data]     /usr/local/share/nltk_data...
[nltk_data]   Unzipping corpora/gutenberg.zip.
True

Example: Switch default data directory

The current data directory is ~/nltk_data and one would like to use the directory /usr/local/share/nltk_data instead.

Move the data directory and add NLTK_DATA to your shell profile.

$ mv ~/nltk_data /usr/local/share/nltk_data
$ echo "export NLTK_DATA=/usr/local/share/nltk_data" >> ~/.bashrc
$ source ~/.bashrc
$ echo $NLTK_DATA
/usr/local/share/nltk_data

Now nltk uses /usr/local/share/nltk_data as defined in NLTK_DATA.

$ python
Python 3.10.6 (main, Sep  5 2022, 11:08:58) [Clang 11.0.0 (clang-1100.0.33.17)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import nltk
>>> nltk.download('gutenberg')
[nltk_data] Downloading package gutenberg to
[nltk_data]     /usr/local/share/nltk_data...
[nltk_data]   Package gutenberg is already up-to-date!
True

score 0 · Answer 7 · answered Jun 28 '23 at 13:08

smci said:

nltk seems to totally ignore its own environment variable NLTK_DATA

Actually,

nltk ignores NLTK_DATA for nltk.download(...)
nltk ignores NLTK_DATA defined after 'import nltk'

so :

import  nltk
os.environ["NLTK_DATA"]="/some/path"
// use nltk and expect it to use "/some/path"

woulds not work, whereas :

os.environ["NLTK_DATA"]="/some/path"
import  nltk
// use nltk and expect it to use "/some/path"

would work.

Furthermore, for nltk.download, you can define your selected download_dir

Change nltk.download() path directory from default ~/ntlk_data

7 Answers7

Command line installation

tl;dr

Explanation

Example: Create new default data directory

Example: Switch default data directory

Linked