Tasks and Datasets in ParlAI

Authors: Alexander Holden Miller, Filipe de Avila Belbute Peres, Jason Weston, Emily Dinan

ParlAI can support fixed dialogue data for supervised learning (which we call a dataset) or even dynamic tasks involving an environment, agents and possibly rewards (we refer to the general case as a task).

In this tutorial we will go over the different ways a new task (or dataset) can be created.

All setups are handled in pretty much the same way, with the same API, but there are less steps of course to make a basic dataset.

For a fast way to add a new dataset, go to the Quickstart below.

For more complete instructions, or a more complicated setup (like streaming large data), go to the section Creating a new task: the more complete way.

Quickstart: Adding a new dataset

Let’s look at the easiest way of getting a new dataset into ParlAI first.

If you have a dialogue, QA or other text-only dataset that you can put in a text file in the format (called ParlAI Dialog Format) we will now describe, you can just load it directly from there, with no extra code!

Here’s an example dataset with a single episode with 2 examples:

text:hello how are you today?   labels:i'm great thanks! what are you doing?
text:i've just been biking. labels:oh nice, i haven't got on a bike in years!   episode_done:True

Suppose that data is in the file /tmp/data.txt


There are tabs between each field above which are rendered in the browser as four spaces. Be sure to change them to tabs for the command below to work.

We could look at that data using the usual display data script:

python parlai/scripts/display_data.py -t fromfile:parlaiformat --fromfile_datapath /tmp/data.txt
<.. snip ..>
[creating task(s): fromfile:parlaiformat]
[loading parlAI text data:/tmp/data.txt]
[/tmp/data.txt]: hello how are you today?
[labels: i'm great thanks! what are you doing?]
   [RepeatLabelAgent]: i'm great thanks! what are you doing?
[/tmp/data.txt]: i've just been biking.
[labels: oh nice, i haven't got on a bike in years!]
   [RepeatLabelAgent]: oh nice, i haven't got on a bike in years!
- - - - - - - - - - - - - - - - - - - - -
[ loaded 1 episodes with a total of 2 examples ]

The text file data format is called ParlAI Dialog format, and is described in the teachers documentation; and in pyparlai.core.teachers.ParlAIDialogTeacher. Essentially, there is one training example every line, and each field in a ParlAI message is tab separated with the name of the field, followed by a colon. E.g. the usual fields like ‘text’, ‘labels’, ‘label_candidates’ etc. can all be used, or you can add your own fields too if you have a special use for them.


Data folds are not automatically generated. Using fromfile as above will result in the same data used for train, validation and test. See the next section on how to have separate folds.

Handling Separate Train/Valid/Test data

Once you’ve gotten the basics of a data working above, you might want to separate out the data into specific train/valid/test sets, as the above example uses the same data for all folds. This is also easy to do. Simply separate the data into three separate files: mydata_train.txt, mydata_valid.txt and mydata_test.txt. Afterwards, modify your parlai call as follows:

python parlai/scripts/display_data.py -t fromfile:parlaiformat --fromfile-datapath /tmp/mydata --fromfile-datatype-extension true

This will cause the system to add the _train.txt, _valid.txt, and _test.txt suffixes at the appropriate times during training, evaluation, etc.

Json file format (instead of text file format)

Prefer to use json instead of text files? No problem, the setup is almost the same! Make a file like this instead (using the same example data as above):

{ "dialog": [ [  {"id": "partner1", "text": "hello how are you today?"},  {"id": "partner2", "text": "i'm great thanks! what are you doing?"},  {"id": "partner1", "text": "i've just been bikinig."},        {"id": "partner2", "text": "oh nice, i haven't got on a bike in years!"} ] ]}

We can then again look at that data using the usual display data script, using the jsonfile teacher:

python parlai/scripts/display_data.py -t jsonfile --json-datapath /tmp/data.json
<.. snip ..>
[creating task(s): jsonfile]
[loading data from json file into task:/tmp/data.json]
- - - NEW EPISODE: tmp/data.json - - -
hello how are you today?
   i'm great thanks! what are you doing?
i've just been biking.
   oh nice, i haven't got on a bike in years!
[epoch done]
[loaded 1 episodes with a total of 2 examples]

The file format consists of one dialogue episode per line, and closely follows the ParlAI messages format. See here for more documentation.

For train/valid/test splits, you can do the same as for text files, using the analogous –jsonfile-datatype-extension true flag.

Creating a new task: the more complete way

Of course after your first hacking around you may want to actually check this code in so that you can share it with others!

Tasks code is located in the parlai/tasks directory.

You will need to create a directory for your new task there.

If your data is in the ParlAI format, you effectively only need a tiny bit of boilerplate to load it, see e.g. the code for the fromfile task agent we just used.

But right now, let’s go through all the steps. You will need to:

  1. Add an __init__.py file to make sure imports work correctly.

  2. Implement build.py to download and build any needed data (see Part 1: Building the Data).

  3. Implement agents.py, with at least a DefaultTeacher which extends Teacher or one of its children (see Part 2: Creating the Teacher).

  4. Add the task to the the task list (see Part 3: Add Task to Task List).

Below we go into more details for each of these steps.

Part 1: Building the Data


If you do not intend to commit your task to ParlAI, and instead wish to load your data locally from disk for your own purposes, you can skip this section and go straight to Part 2.

We first need to create functionality for downloading and setting up the dataset that is going to be used for the task. This is done in the build.py file. Useful functionality for setting up data can be found in parlai.core.build_data.

import parlai.core.build_data as build_data
import os

Now we define our build method, which takes in the argument opt, which contains parsed arguments from the command line (or their default), including the path to the data directory. We can also define a version string, so that the data is removed and updated automatically for other ParlAI users if we make changes to this task (here it was just left it as None). We then use the build_data utilities to check if this data hasn’t been previously built or if the version is outdated. If not, we proceed to creating the directory for the data, and then downloading and uncompressing it. Finally, we mark the build as done, so that build_data.built() returns true from now on. Below is an example of setting up the SQuAD dataset.

        '<checksum for this file>',
        '<checksum for this file>',

def build(opt):
    dpath = os.path.join(opt['datapath'], 'SQuAD')
    version = None

    if not build_data.built(dpath, version_string=version):
        print('[building data: ' + dpath + ']')
        if build_data.built(dpath):
            # An older version exists, so remove these outdated files.

        # Download the data.
        for downloadable_file in RESOURCES[:2]:

        # Mark the data as built.
        build_data.mark_done(dpath, version_string=version)

Part 2: Creating the Teacher

Now that we have our data, we need an agent that understand the task’s structure and is able to present it. In other words, we need a Teacher. Every task requires an agents.py file in which we define the agents for the task. It is there that we will define our teacher.

Which base teacher should I use?

We will describe three possible teachers to subclass; you can choose one of them based on your needs:

  1. ParlAIDialogTeacher: This is the simplest method available, and expects to load a text file of data in ParlAI Dialog format (described above). More details are shown in the section ParlAIDialogTeacher.

  2. DialogTeacher: If the data is not in ParlAI Dialog format, one can still use the DialogTeacher which automates much of the work in setting up a dialog task, but gives the user more flexibility in loading the data from the disk. This is shown in the section DialogTeacher.

  3. ChunkTeacher: Use this teacher if you have a large dataset that cannot fit in memory at once. In the ChunkTeacher section, we show how to break the dataset into smaller “chunks” that are efficiently loaded.

Finally, if the requirements for the task do not fit any of the above, one can still write a task from scratch without much trouble. This is shown in the section Task from Scratch. For example, a dynamic task which adjusts its response based on the received input rather than using fixed logs is better suited to this approach.


For this class, the user must implement at least an __init__() function, and often only that.

In this section we will illustrate the process of using the ParlAIDialogTeacher class by adding the Twitter dataset. This task has data in textual form and has been formatted to follow the ParlAI Dialog format. It is thus very simple to implement it using ParlAIDialogTeacher. More information on this class and the dialog format can be found in the teachers documentation <core/teachers>.

In this task, the agent is presented with questions about movies that are answerable from Wikipedia. A sample dialog is demonstrated below.

[twitter]: burton is a fave of mine,even his average films are better than most directors.
[labels: keeping my fingers crossed that he still has another ed wood in him before he retires.]
- - - - - - - - - - - - - - - - - - - - -
[twitter]: i saw someone say that we should use glass straws..
[labels: glass or paper straws - preferably no 'straw' waste. ban !]

Every task requires a DefaultTeacher. Since we are subclassing ParlAIDialogTeacher, we only have to initialize the class and set a few option parameters, as shown below.

class DefaultTeacher(ParlAIDialogTeacher):
    def __init__(self, opt, shared=None):
        opt = copy.deepcopy(opt)

        # get datafile
        opt['parlaidialogteacher_datafile'] = _path(opt, '')

        super().__init__(opt, shared)

We can notice there was a call to a _path() method, which returns the path to the correct datafile. The path to the file is then stored in the options dictionary under the parlaidialogteacher_datafile key. This item is passed to setup_data() so that subclasses can just override the path instead of the function. We still need to implement this _path() method. The version for this example is presented below. It first ensures the data is built by calling the build() method described in Part 1. It then sets up the paths for the built data.


Note again, that if you are loading data locally from disk, you can skip the call to build here, and instead simply return the path to your data file locally given opt['datatype'].

from .build import build

def _path(opt, filtered):
    # build the data if it does not exist

    # set up path to data (specific to each dataset)
    dt = opt['datatype'].split(':')[0]
    return os.path.join(opt['datapath'], 'Twitter', dt + '.txt')

And this is all that needs to be done to create a teacher for our task using ParlAIDialogTeacher.

To access this data, we can now use the display_data.py file in the examples directory:

parlai display_data -t twitter


For this class, the user must also implement their own setup_data() function, but the rest of the work of supporting hogwild or batching, streaming data from disk, processing images, and more is taken care of for them.

In this section we will demonstrate the process of using the DialogTeacher class by adding the Stanford Question Answering Dataset (SQuAD) dataset. The data on disk downloaded from the SQuAD website does not fit the basic ParlAIDialogTeacher format described above. Still, using DialogTeacher makes it easy to implement dialog tasks such as this one.

In this task, the agent is presented with a paragraph from Wikipedia and asked to answer a question about it.

[id]: squad
[text]: In October 2014, Beyoncé signed a deal to launch an activewear line of clothing with British fashion retailer Topshop. The 50-50 venture is called Parkwood Topshop Athletic Ltd and is scheduled to launch its first dance, fitness and sports ranges in autumn 2015. The line will launch in April 2016.
When will the full line appear?

[labels]: April 2016

We will call our teacher SquadTeacher. Let’s initialize this class first.

class SquadTeacher(DialogTeacher):
    def __init__(self, opt, shared=None):
        self.datatype = opt['datatype']
        build(opt)  # NOTE: the call to build here
        suffix = 'train' if opt['datatype'].startswith('train') else 'dev'
        # whatever is placed into datafile will be passed as the argument to
        # setup_data in the next section.
        opt['datafile'] = os.path.join(opt['datapath'], 'SQuAD', suffix + '-v1.1.json')
        self.id = 'squad'
        super().__init__(opt, shared)

The id field names the teacher in the dialog.

By creating SquadTeacher as a subclass of DialogTeacher, the job of creating a teacher for this task becomes much simpler: most of the work that needs to be done will limit itself to defining a setup_data method. This method is a generator that will take in a path to the data and yield a pair of elements for each call. The first element of the pair is a dictionary containing a dialogue act (with required fields text and labels and any other additional field required by your task, such as label_candidates). In this case, we only return text and labels. The second is a boolean flag new_episode? which indicates if the current query starts a new episode or not.

More information on this format can be found in the documentation under DialogData in the teachers documentation <core/teachers> (setup_data is provided as a data_loader to DialogData).

The sample setup_data method for our task is presented below.

def setup_data(self, path):
    # note that path is the value provided by opt['datafile']
    print('loading: ' + path)
    with PathManager.open(path) as data_file:
        self.squad = json.load(data_file)['data']
    for article in self.squad:
        # each paragraph is a context for the attached questions
        for paragraph in article['paragraphs']:
            # each question is an example
            for qa in paragraph['qas']:
                question = qa['question']
                answers = tuple(a['text'] for a in qa['answers'])
                context = paragraph['context']
                yield {"text": content + "\n" + question, "labels": answers}, True

As we can see from the code above, for this task, each episode consists of only one query, thus new_episode? is always true (i.e., each query is the start of its episode). This could also vary depending on the task.

Finally, one might notice that no reward, label candidates, or a path to an image were provided in the tuple (all are set to None). These fields are not relevant to this task.

The only thing left to be done for this part is to define a DefaultTeacher class. This is a requirement for any task, as the create_agent method looks for a teacher named this. We can simply default to the class we have built so far.

class DefaultTeacher(SquadTeacher):

And we have finished building our task.

Chunk Teacher

Chunk Teacher is useful for streaming large amounts of data (read: does not fit into memory), that naturally separate into several separate files (or chunks). The data is separated into chunks and loaded one chunk at a time. Loads the data off of the main thread.

To implement a chunk teacher, you have to write the following functions:

  • _get_data_folder: Returns the path to the directory containing the chunks.

  • get_num_samples: Given opt, returns a tuple containing (num_episodes, num_examples). Since we are streaming this data, we must know the number of examples a priori.

  • get_fold_chunks: Given opt, returns a list of chunks indices. For example, we might separate the chunks based on the data split, given by opt['datatype'].

  • load_from_chunk: Given a chunk index, loads the associated file and returns a list of samples.

  • create_message: Given a single sample item from the list returned by load_from_chunk, create a Message to return.

We create an example teacher to demonstrate. Let’s suppose that /tmp/path_to_my_chunks/ is the directory containing our chunks, and each chunk file (e.g. /tmp/path_to_my_chunks/1.txt) has the following format:

<input 1>\t<output 1>
<input 2>\t<output 2>
<input 3>\t<output 3>
<input 100>\t<output 100>

Then our chunk teacher would look like the following:

class ExampleChunkTeacher(ChunkTeacher):
    def _get_data_folder(self):
        # return the path to directory containing your chunks
        return '/tmp/path_to_my_chunks/'

    def get_num_samples(self, opt) -> Tuple[int, int]:
        # return the number of episodes and examples
        # in this case, all of our episodes are single examples
        # so they are the same number
        datatype = opt['datatype']
        if 'train' in datatype:
            return 300, 300  # each chunk contains 100 examples
        elif 'valid' in datatype:
            return 100, 100
        elif 'test' in datatype:
            return 100, 100

    def get_fold_chunks(self, opt) -> List[int]:
        # in this case, our train split contains 3 chunks and
        # valid and test each contain 1
        datatype = opt['datatype']
        if 'train' in datatype:
            return [1, 2, 3]
        elif 'valid' in datatype:
            return [4]
        elif 'test' in datatype:
            return [5]

    def load_from_chunk(self, chunk_idx: int):
        # we load the chunk specified by chunk_idx and return a
        # list of outputs
        chunk_path = os.path.join(self._get_data_folder(), f'{chunk_idx}.txt')
        output = []
        with open(chunk_path, 'r') as f:
            for line in f.readlines():
                txt_input, txt_output = line.split('\t')
                output.append((txt_input, txt_output))

        return output

    def create_message(self, sample_item, entry_idx=0):
        # finally, we return a message given an element from the list
        # returned by `load_from_chunk`
        text, label = sample_item
        return {'id': 'Example Chunk Teacher', 'text': text, 'labels': [label], 'episode_done': True}


Chunk Teacher only works with streaming data, so make sure to run with -dt train:stream (or -dt valid:stream or -dt test:stream) when using your chunk data.

Task from Scratch

In this case, one would inherit from the Teacher class. For this class, at least the act() method and probably the observe() method must be overriden. Quite a bit of functinoality will not be built in, such as a support for hogwild and batching, but metrics will be set up and can be used to track stats like the number of correctly answered examples.

In general, extending Teacher directly is not recommended unless the above classes definitely do not fit your task. We still have a few remnants which do this in our code base instead of using FixedDialogTeacher, but this will require one to do extra work to support batching and hogwild if desired.

However, extending teacher directly is necessary for non-fixed data. For example, one might have a like the full negotiation version of the dealnodeal task, where episodes are variable-length (it continues until one player ends the discussion).

In this case, just implement the observe() function to handle seeing the text from the other agent, and the act() function to take an action (such as sending text or other fields such as reward to the other agent).

Part 3: Add Task to Task List

Now that our task is complete, we must add an entry to the task_list.py file in parlai/tasks. This file just contains a json-formatted list of all tasks, with each task represented as a dictionary. Sample entries for our tasks are presented below.

    # other tasks...
        "id": "MTurkWikiMovies",
        "display_name": "MTurk WikiMovies",
        "task": "mturkwikimovies",
        "tags": [ "all",  "QA" ],
        "description": "Closed-domain QA dataset asking MTurk-derived questions about movies, answerable from Wikipedia. From Li et al. '16. Link: https://arxiv.org/abs/1611.09823"
        "id": "MNIST_QA",
        "display_name": "MNIST_QA",
        "task": "mnist_qa",
        "tags": [ "all", "Visual" ],
        "description": "Task which requires agents to identify which number they are seeing. From the MNIST dataset."
        "id": "VQAv2",
        "display_name": "VQAv2",
        "task": "vqa_v2",
        "tags": [ "all", "Visual" ],
        "description": "Bigger, more balanced version of the original VQA dataset. From Goyal et al. '16. Link: https://arxiv.org/abs/1612.00837"
    # other tasks...

Part 4: Executing the Task

A simple way of testing the basic functionality in a task is to run the display_data.py example in the examples directory. Now that the work is done, we can pass it to ParlAI by using the -t flag. For example, to execute the MTurk WikiMovies task we should call:

python display_data.py -t mturkwikimovies

To run the MNIST_QA task, while displaying the images in ascii format, we could call:

python display_data.py -t mnist_qa -im ascii

And for VQAv2:

python display_data.py -t vqa_v2

Part 5: Contributing upstream

Wow! Now that you’ve created your task, perhaps you should add it to ParlAI! We welcome new datasets from all contributors.

If you do want to make a pull request, we will usually expect that you submit a Regression Test with your dataset. The regression test is a sort of automatic test which ensures your teacher gives consistent output in different versions of ParlAI.

Adding a regression test is easy. Just add the following “test.py” into the same folder as your tasks.

#!/usr/bin/env python3

# Copyright (c) Facebook, Inc. and its affiliates.
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.

from parlai.utils.testing import AutoTeacherTest  # noqa: F401

class TestDefaultTeacher(AutoTeacherTest):
    task = 'myteacher'  # replace with your teacher name

Next, run your test. It will always fail the first time.

pytest parlai/tasks/myteacher/test.py

You will see many error messages about “Failed: File not found in data directory, created.” This is expected the first time. You should now see a “test” folder appear in the same directory, containing several .yml files. Add these files to your git commit.

Next run the test again. This time, you should see all tests pass. If so, go ahead and create your PR.

Note, if you need to make further changes to your teacher, you may need to update the regression fixtures. You can do this by adding --force-regen to the pytest arguments:

pytest --force-regen parlai/tasks/myteacher/test.py

Similar to the first time you ran the test, you may see failure. Rerun again to ensure things pass.