Dataframe as datasource in torchtext
Asked Answered
J

2

9

I have a dataframe, which has two columns (review and sentiment). I am using pytorch and torchtext library for preprocessing data. Is it possible to use dataframe as source to read data from, in torchtext? I am looking for something similar to, but not

data.TabularDataset.splits(path='./data')

I have performed some operation (clean, change to required format) on data and final data is in a dataframe.

If not torchtext, what other package would you suggest that would help in preprocessing text data present in a datarame. I could not find anything online. Any help would be great.

Juliannejuliano answered 2/10, 2018 at 4:33 Comment(0)
W
15

Adapting the Dataset and Example classes from torchtext.data

    from torchtext.data import Field, Dataset, Example
    import pandas as pd

     class DataFrameDataset(Dataset):
         """Class for using pandas DataFrames as a datasource"""
         def __init__(self, examples, fields, filter_pred=None):
             """
             Create a dataset from a pandas dataframe of examples and Fields
             Arguments:
                 examples pd.DataFrame: DataFrame of examples
                 fields {str: Field}: The Fields to use in this tuple. The
                     string is a field name, and the Field is the associated field.
                 filter_pred (callable or None): use only exanples for which
                     filter_pred(example) is true, or use all examples if None.
                     Default is None
             """
             self.examples = examples.apply(SeriesExample.fromSeries, args=(fields,), axis=1).tolist()
             if filter_pred is not None:
                 self.examples = filter(filter_pred, self.examples)
             self.fields = dict(fields)
             # Unpack field tuples
             for n, f in list(self.fields.items()):
                 if isinstance(n, tuple):
                     self.fields.update(zip(n, f))
                     del self.fields[n]

     class SeriesExample(Example):
         """Class to convert a pandas Series to an Example"""
        
         @classmethod
         def fromSeries(cls, data, fields):
             return cls.fromdict(data.to_dict(), fields)

         @classmethod
         def fromdict(cls, data, fields):
             ex = cls()
             
             for key, field in fields.items():
                 if key not in data:
                     raise ValueError("Specified key {} was not found in "
                     "the input data".format(key))
                 if field is not None:
                     setattr(ex, key, field.preprocess(data[key]))
                 else:
                     setattr(ex, key, data[key])
             return ex

Then, first define fields using torchtext.data fields. For example:

    TEXT = data.Field(tokenize='spacy')
    LABEL = data.LabelField(dtype=torch.float)
    TEXT.build_vocab(train, max_size=25000, vectors="glove.6B.100d") 
    LABEL.build_vocab(train)
    fields = { 'sentiment' : LABEL, 'review' : TEXT }

before simply loading the dataframes:

    train_ds = DataFrameDataset(train_df, fields)
    valid_ds = DataFrameDataset(valid_df, fields)
Wrangler answered 4/10, 2018 at 22:1 Comment(9)
I have tried implementing this, but it is not clear what "fields" should consist of or how it is constructed. In the questions case with two "Keys" in the dataframe: review and sentiment. Any further elaboration would highly appreciatedReames
Figured it out, it should be in the format of a dictionary where each key is series name and each value is what to do them: fields = { 'sentiment' : LABEL, 'review' : TEXT } where label and text are torchtext data fields such as: TEXT = data.Field(tokenize='spacy') LABEL = data.LabelField(dtype=torch.float) TEXT.build_vocab(train, max_size=25000, vectors="glove.6B.100d") LABEL.build_vocab(train)Reames
+1 because this implementation follows the original implementation logic and style pytorch.org/text/_modules/torchtext/data/…Disk
@NicolaiF, I just edited the answer to reflect your comment and make it easier for the readersWrangler
@GeoffreyNegiar: The last statement return ex doesn't seem indented properly. It's not clear to me at what indentation level it should be.Breazeale
@Reames : What does variable 'train' refer to in the line: TEXT.build_vocab(train, max_size=25000, vectors="glove.6B.100d") LABEL.build_vocab(train) ?Specialism
@Breazeale I think it is: the Example is returned once all the fields are processed.Wrangler
@JohnHawkins I'm not sure I remember what it was... Possibly train_df.Wrangler
How can I sort the values inside the fields object?... When I run LABEL.vocab.stoi I receive back defaultdict(None, {0: 0, 2: 1, 1: 2}) while it should return defaultdict(None, {0: 0, 1: 1, 2: 2})Tripe
S
0

Thanks Geoffrey.

From looking at the source code for torchtext.data.field

https://pytorch.org/text/_modules/torchtext/data/field.html

It looks like the 'train' parameter needs to be either a Dataset already, or some iterable source of text data. But given we haven't created a dataset at this point I am guessing you have passed in just the column of text from the dataframe.

Specialism answered 11/10, 2020 at 9:30 Comment(0)

© 2022 - 2024 — McMap. All rights reserved.