Functions and transforms to help gather text data in a `Datasets`

Numericalizing

Numericalization is the step in which we convert tokens to integers. The first step is to build a correspondence token to index that is called a vocab.

make_vocab[source]

make_vocab(count, min_freq=3, max_vocab=60000, special_toks=None)

Create a vocab of max_vocab size from Counter count with items present more than min_freq

If there are more than max_vocab tokens, the ones kept are the most frequent.

count = Counter(['a', 'a', 'a', 'a', 'b', 'b', 'c', 'c', 'd'])
test_eq(set([x for x in make_vocab(count) if not x.startswith('xxfake')]), 
        set(defaults.text_spec_tok + 'a'.split()))
test_eq(len(make_vocab(count))%8, 0)
test_eq(set([x for x in make_vocab(count, min_freq=1) if not x.startswith('xxfake')]), 
        set(defaults.text_spec_tok + 'a b c d'.split()))
test_eq(set([x for x in make_vocab(count,max_vocab=12, min_freq=1) if not x.startswith('xxfake')]), 
        set(defaults.text_spec_tok + 'a b c'.split()))

class TensorText[source]

TensorText(x, **kwargs) :: TensorBase

Semantic type for a tensor representing text in language modeling

class LMTensorText[source]

LMTensorText(x, **kwargs) :: TensorText

Semantic type for a tensor representing text in language modeling

class Numericalize[source]

Numericalize(vocab=None, min_freq=3, max_vocab=60000, special_toks=None, pad_tok=None) :: Transform

Reversible transform of tokenized texts to numericalized ids

If no vocab is passed, one is created at setup from the data, using make_vocab with min_freq and max_vocab.

start = 'This is an example of text'
num = Numericalize(min_freq=1)
num.setup(L(start.split(), 'this is another text'.split()))
test_eq(set([x for x in num.vocab if not x.startswith('xxfake')]), 
        set(defaults.text_spec_tok + 'This is an example of text this another'.split()))
test_eq(len(num.vocab)%8, 0)
t = num(start.split())

test_eq(t, tensor([11, 9, 12, 13, 14, 10]))
test_eq(num.decode(t), start.split())
num = Numericalize(min_freq=2)
num.setup(L('This is an example of text'.split(), 'this is another text'.split()))
test_eq(set([x for x in num.vocab if not x.startswith('xxfake')]), 
        set(defaults.text_spec_tok + 'is text'.split()))
test_eq(len(num.vocab)%8, 0)
t = num(start.split())
test_eq(t, tensor([0, 9, 0, 0, 0, 10]))
test_eq(num.decode(t), f'{UNK} is {UNK} {UNK} {UNK} text'.split())

class LMDataLoader[source]

LMDataLoader(dataset, lens=None, cache=2, bs=64, seq_len=72, num_workers=0, shuffle=False, verbose=False, do_setup=True, pin_memory=False, timeout=0, batch_size=None, drop_last=False, indexed=None, n=None, device=None, wif=None, before_iter=None, after_item=None, before_batch=None, after_batch=None, after_iter=None, create_batches=None, create_item=None, create_batch=None, retain=None, get_idxs=None, sample=None, shuffle_fn=None, do_batch=None) :: TfmdDL

A DataLoader suitable for language modeling

dataset should be a collection of numericalized texts for this to work. lens can be passed for optimizing the creation, otherwise, the LMDataLoader will do a full pass of the dataset to compute them. cache is used to avoid reloading items unnecessarily.

The LMDataLoader will concatenate all texts (maybe shuffled) in one big stream, split it in bs contiguous sentences, then go through those seq_len at a time.

bs,sl = 4,3
ints = L([0,1,2,3,4],[5,6,7,8,9,10],[11,12,13,14,15,16,17,18],[19,20],[21,22,23],[24]).map(tensor)
dl = LMDataLoader(ints, bs=bs, seq_len=sl)
test_eq(list(dl),
    [[tensor([[0, 1, 2], [6, 7, 8], [12, 13, 14], [18, 19, 20]]),
      tensor([[1, 2, 3], [7, 8, 9], [13, 14, 15], [19, 20, 21]])],
     [tensor([[3, 4, 5], [ 9, 10, 11], [15, 16, 17], [21, 22, 23]]),
      tensor([[4, 5, 6], [10, 11, 12], [16, 17, 18], [22, 23, 24]])]])
dl = LMDataLoader(ints, bs=bs, seq_len=sl, shuffle=True)
for x,y in dl: test_eq(x[:,1:], y[:,:-1])
((x0,y0), (x1,y1)) = tuple(dl)
#Second batch begins where first batch ended
test_eq(y0[:,-1], x1[:,0]) 
test_eq(type(x0), LMTensorText)

Classification

For classification, we deal with the fact that texts don't all have the same length by using padding.

pad_input[source]

pad_input(samples, pad_idx=1, pad_fields=0, pad_first=False, backwards=False)

Function that collect samples and adds padding

pad_idx is used for the padding, and the padding is applied to the pad_fields of the samples. The padding is applied at the beginning if pad_first is True, and if backwards is added, the tensors are flipped.

test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0), 
        [(tensor([1,2,3]),1), (tensor([4,5,0]),2), (tensor([6,0,0]), 3)])
test_eq(pad_input([(tensor([1,2,3]), (tensor([6]))), (tensor([4,5]), tensor([4,5])), (tensor([6]), (tensor([1,2,3])))], pad_idx=0, pad_fields=1), 
        [(tensor([1,2,3]),(tensor([6,0,0]))), (tensor([4,5]),tensor([4,5,0])), ((tensor([6]),tensor([1, 2, 3])))])
test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0, pad_first=True), 
        [(tensor([1,2,3]),1), (tensor([0,4,5]),2), (tensor([0,0,6]), 3)])
test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0, backwards=True), 
        [(tensor([3,2,1]),1), (tensor([5,4,0]),2), (tensor([6,0,0]), 3)])
x = test_eq(pad_input([(tensor([1,2,3]),1), (tensor([4,5]), 2), (tensor([6]), 3)], pad_idx=0, backwards=True), 
        [(tensor([3,2,1]),1), (tensor([5,4,0]),2), (tensor([6,0,0]), 3)])

pad_input_chunk[source]

pad_input_chunk(samples, pad_idx=1, pad_first=True, seq_len=72)

Pad samples by adding padding by chunks of size seq_len

The difference with the base pad_input is that most of the padding is applied first (if pad_first=True) or at the end (if pad_first=False) but only by a round multiple of seq_len. The rest of the padding is applied to the end (or the beginning if pad_first=False). This is to work with SequenceEncoder with recurrent models.

test_eq(pad_input_chunk([(tensor([1,2,3,4,5,6]),1), (tensor([1,2,3]), 2), (tensor([1,2]), 3)], pad_idx=0, seq_len=2), 
        [(tensor([1,2,3,4,5,6]),1), (tensor([0,0,1,2,3,0]),2), (tensor([0,0,0,0,1,2]), 3)])
test_eq(pad_input_chunk([(tensor([1,2,3,4,5,6]),), (tensor([1,2,3]),), (tensor([1,2]),)], pad_idx=0, seq_len=2), 
        [(tensor([1,2,3,4,5,6]),), (tensor([0,0,1,2,3,0]),), (tensor([0,0,0,0,1,2]),)])
test_eq(pad_input_chunk([(tensor([1,2,3,4,5,6]),), (tensor([1,2,3]),), (tensor([1,2]),)], pad_idx=0, seq_len=2, pad_first=False), 
        [(tensor([1,2,3,4,5,6]),), (tensor([1,2,3,0,0,0]),), (tensor([1,2,0,0,0,0]),)])

class SortedDL[source]

SortedDL(dataset, sort_func=None, res=None, bs=64, shuffle=False, num_workers=None, verbose=False, do_setup=True, pin_memory=False, timeout=0, batch_size=None, drop_last=False, indexed=None, n=None, device=None, wif=None, before_iter=None, after_item=None, before_batch=None, after_batch=None, after_iter=None, create_batches=None, create_item=None, create_batch=None, retain=None, get_idxs=None, sample=None, shuffle_fn=None, do_batch=None) :: TfmdDL

A DataLoader that goes throught the item in the order given by sort_func

res is the result of sort_func applied on all elements of the dataset. You can pass it if available to make the init faster by avoiding an initial pass over the whole dataset. If shuffle is True, this will shuffle a bit the results of the sort to have items of roughly the same size in batches, but not in the exact sorted order.

ds = [(tensor([1,2]),1), (tensor([3,4,5,6]),2), (tensor([7]),3), (tensor([8,9,10]),4)]
dl = SortedDL(ds, bs=2, before_batch=partial(pad_input, pad_idx=0))
test_eq(list(dl), [(tensor([[ 3,  4,  5,  6], [ 8,  9, 10,  0]]), tensor([2, 4])), 
                   (tensor([[1, 2], [7, 0]]), tensor([1, 3]))])
ds = [(tensor(range(random.randint(1,10))),i) for i in range(101)]
dl = SortedDL(ds, bs=2, create_batch=partial(pad_input, pad_idx=-1), shuffle=True, num_workers=0)
batches = list(dl)
max_len = len(batches[0][0])
for b in batches: 
    assert(len(b[0])) <= max_len 
    test_ne(b[0][-1], -1)

TransformBlock for text

To use the data block API, you will need this build block for texts.

class TextBlock[source]

TextBlock(tok_tfm, vocab=None, is_lm=False, seq_len=72, min_freq=3, max_vocab=60000, special_toks=None, pad_tok=None) :: TransformBlock

A TransformBlock for texts

For efficient tokenization, you probably want to use one of the factory methods. Otherwise, you can pass your custom tok_tfm that will deal with tokenization (if your texts are already tokenized, you can pass noop), a vocab, or leave it to be inferred on the texts using min_freq and max_vocab.

is_lm indicates if we want to use texts for language modeling or another task, seq_len is only necessary to tune if is_lm=False, and is passed along to pad_input_chunk.

TextBlock.from_df[source]

TextBlock.from_df(text_cols, vocab=None, is_lm=False, seq_len=72, min_freq=3, max_vocab=60000, tok_func='SpacyTokenizer', rules=None, sep=' ', n_workers=56, mark_fields=None, res_col_name='text', **kwargs)

Build a TextBlock from a dataframe using text_cols

vocab, is_lm, seq_len, min_freq and max_vocab are passed to the main init, the other argument to Tokenizer.from_df.

TextBlock.from_folder[source]

TextBlock.from_folder(path, vocab=None, is_lm=False, seq_len=72, min_freq=3, max_vocab=60000, tok_func='SpacyTokenizer', rules=None, extensions=None, folders=None, output_dir=None, n_workers=56, encoding='utf8', **kwargs)

Build a TextBlock from a path

vocab, is_lm, seq_len, min_freq and max_vocab are passed to the main init, the other argument to Tokenizer.from_folder.

class TextDataLoaders[source]

TextDataLoaders(*loaders, path='.', device=None) :: DataLoaders

Basic wrapper around several DataLoaders with factory methods for NLP problems

You should not use the init directly but one of the following factory methods. All those factory methods accept as arguments:

  • text_vocab: the vocabulary used for numericalizing texts (if not passed, it's infered from the data)
  • tok_tfm: if passed, uses this tok_tfm instead of the default
  • seq_len: the sequence length used for batch
  • bs: the batch size
  • val_bs: the batch size for the validation DataLoader (defaults to bs)
  • shuffle_train: if we shuffle the training DataLoader or not
  • device: the PyTorch device to use (defaults to default_device())

TextDataLoaders.from_folder[source]

TextDataLoaders.from_folder(path, train='train', valid='valid', valid_pct=None, seed=None, vocab=None, text_vocab=None, is_lm=False, tok_tfm=None, seq_len=72, bs=64, val_bs=None, shuffle_train=True, device=None)

Create from imagenet style dataset in path with train and valid subfolders (or provide valid_pct)

If valid_pct is provided, a random split is performed (with an optional seed) by setting aside that percentage of the data for the validation set (instead of looking at the grandparents folder). If a vocab is passed, only the folders with names in vocab are kept.

Here is an example on a sample of the IMDB movie review dataset:

path = untar_data(URLs.IMDB)
dls = TextDataLoaders.from_folder(path)
dls.show_batch(max_n=3)
text category
0 xxbos xxmaj match 1 : xxmaj tag xxmaj team xxmaj table xxmaj match xxmaj bubba xxmaj ray and xxmaj spike xxmaj dudley vs xxmaj eddie xxmaj guerrero and xxmaj chris xxmaj benoit xxmaj bubba xxmaj ray and xxmaj spike xxmaj dudley started things off with a xxmaj tag xxmaj team xxmaj table xxmaj match against xxmaj eddie xxmaj guerrero and xxmaj chris xxmaj benoit . xxmaj according to the rules of the match , both opponents have to go through tables in order to get the win . xxmaj benoit and xxmaj guerrero heated up early on by taking turns hammering first xxmaj spike and then xxmaj bubba xxmaj ray . a xxmaj german xxunk by xxmaj benoit to xxmaj bubba took the wind out of the xxmaj dudley brother . xxmaj spike tried to help his brother , but the referee restrained him while xxmaj benoit and xxmaj guerrero pos
1 xxbos xxmaj by now you 've probably heard a bit about the new xxmaj disney dub of xxmaj miyazaki 's classic film , xxmaj laputa : xxmaj castle xxmaj in xxmaj the xxmaj sky . xxmaj during late summer of 1998 , xxmaj disney released " kiki 's xxmaj delivery xxmaj service " on video which included a preview of the xxmaj laputa dub saying it was due out in " 1 xxrep 3 9 " . xxmaj it 's obviously way past that year now , but the dub has been finally completed . xxmaj and it 's not " laputa : xxmaj castle xxmaj in xxmaj the xxmaj sky " , just " castle xxmaj in xxmaj the xxmaj sky " for the dub , since xxmaj laputa is not such a nice word in xxmaj spanish ( even though they use the word xxmaj laputa many times pos
2 xxbos xxmaj titanic directed by xxmaj james xxmaj cameron presents a fictional love story on the historical setting of the xxmaj titanic . xxmaj the plot is simple , xxunk , or not for those who love plots that twist and turn and keep you in suspense . xxmaj the end of the movie can be figured out within minutes of the start of the film , but the love story is an interesting one , however . xxmaj kate xxmaj winslett is wonderful as xxmaj rose , an aristocratic young lady betrothed by xxmaj cal ( billy xxmaj zane ) . xxmaj early on the voyage xxmaj rose meets xxmaj jack ( leonardo dicaprio ) , a lower class artist on his way to xxmaj america after winning his ticket aboard xxmaj titanic in a poker game . xxmaj if he wants something , he goes and gets it pos

TextDataLoaders.from_df[source]

TextDataLoaders.from_df(df, path='.', valid_pct=0.2, seed=None, text_col=0, label_col=1, label_delim=None, y_block=None, text_vocab=None, is_lm=False, valid_col=None, tok_tfm=None, seq_len=72, bs=64, val_bs=None, shuffle_train=True, device=None)

Create from df in path with valid_pct

seed can optionally be passed for reproducibility. text_col, label_col and optionaly valid_col are indices or names of columns for texts/labels and the validation flag. label_delim can be passed for a multi-label problem if your labels are in one column, separated by a particular char. y_block should be passed to indicate your type of targets, in case the library did no infer it properly.

Here are examples on subsets of IMDB:

path = untar_data(URLs.IMDB_SAMPLE)
df = pd.read_csv(path/'texts.csv')
df.head(2)
label text is_valid
0 negative Un-bleeping-believable! Meg Ryan doesn't even look her usual pert lovable self in this, which normally makes me forgive her shallow ticky acting schtick. Hard to believe she was the producer on this dog. Plus Kevin Kline: what kind of suicide trip has his career been on? Whoosh... Banzai!!! Finally this was directed by the guy who did Big Chill? Must be a replay of Jonestown - hollywood style. Wooofff! False
1 positive This is a extremely well-made film. The acting, script and camera-work are all first-rate. The music is good, too, though it is mostly early in the film, when things are still relatively cheery. There are no really superstars in the cast, though several faces will be familiar. The entire cast does an excellent job with the script.<br /><br />But it is hard to watch, because there is no good end to a situation like the one presented. It is now fashionable to blame the British for setting Hindus and Muslims against each other, and then cruelly separating them into two countries. There is som... False
dls = TextDataLoaders.from_df(df, path=path, text_col='text', label_col='label', valid_col='is_valid')
dls.show_batch(max_n=3)
text category
0 xxbos xxmaj raising xxmaj victor xxmaj vargas : a xxmaj review \n\n xxmaj you know , xxmaj raising xxmaj victor xxmaj vargas is like sticking your hands into a big , xxunk bowl of xxunk . xxmaj it 's warm and gooey , but you 're not sure if it feels right . xxmaj try as i might , no matter how warm and gooey xxmaj raising xxmaj victor xxmaj vargas became i was always aware that something did n't quite feel right . xxmaj victor xxmaj vargas suffers from a certain xxunk on the director 's part . xxmaj apparently , the director thought that the ethnic backdrop of a xxmaj latino family on the lower east side , and an xxunk storyline would make the film critic proof . xxmaj he was right , but it did n't fool me . xxmaj raising xxmaj victor xxmaj vargas is negative
1 xxbos xxup the xxup shop xxup around xxup the xxup corner is one of the xxunk and most feel - good romantic comedies ever made . xxmaj there 's just no getting around that , and it 's hard to actually put one 's feeling for this film into words . xxmaj it 's not one of those films that tries too hard , nor does it come up with the xxunk possible scenarios to get the two protagonists together in the end . xxmaj in fact , all its charm is xxunk , contained within the characters and the setting and the plot … which is highly believable to xxunk . xxmaj it 's easy to think that such a love story , as beautiful as any other ever told , * could * happen to you … a feeling you do n't often get from other romantic comedies positive
2 xxbos xxmaj now that xxmaj che(2008 ) has finished its relatively short xxmaj australian cinema run ( extremely limited xxunk screen in xxmaj xxunk , after xxunk ) , i can xxunk join both xxunk of " at xxmaj the xxmaj movies " in taking xxmaj steven xxmaj soderbergh to task . \n\n xxmaj it 's usually satisfying to watch a film director change his style / subject , but xxmaj soderbergh 's most recent stinker , xxmaj the xxmaj girlfriend xxmaj xxunk ) , was also missing a story , so narrative ( and editing ? ) seem to suddenly be xxmaj soderbergh 's main challenge . xxmaj strange , after xxunk years in the business . xxmaj he was probably never much good at narrative , just xxunk it well inside " edgy " projects . \n\n xxmaj none of this excuses him this present , almost diabolical negative
dls = TextDataLoaders.from_df(df, path=path, text_col='text', is_lm=True, valid_col='is_valid')
dls.show_batch(max_n=3)
text text_
0 xxbos i was very excited about seeing this film , xxunk a visual xxunk on the relation of artistic beauty and nature , xxunk the kinds of xxunk the likes of " rivers and xxmaj xxunk . " xxmaj however , that 's not what i received . xxmaj instead , i get a fairly uninspired film about how human industry is bad for nature . xxmaj which is clearly a quite i was very excited about seeing this film , xxunk a visual xxunk on the relation of artistic beauty and nature , xxunk the kinds of xxunk the likes of " rivers and xxmaj xxunk . " xxmaj however , that 's not what i received . xxmaj instead , i get a fairly uninspired film about how human industry is bad for nature . xxmaj which is clearly a quite xxunk
1 each other . xxup the xxup end . xxbos xxmaj hip . xxmaj erotic . xxmaj xxunk sexy … whatever . xxmaj it 's " the xxmaj xxunk " with xxunk . \n\n xxmaj no , seriously . xxmaj the cop saves the girl ( waitress ! ) from the big monster and refers to himself as her ' protector ' . xxmaj the lead actor xxmaj ryan xxmaj xxunk does a other . xxup the xxup end . xxbos xxmaj hip . xxmaj erotic . xxmaj xxunk sexy … whatever . xxmaj it 's " the xxmaj xxunk " with xxunk . \n\n xxmaj no , seriously . xxmaj the cop saves the girl ( waitress ! ) from the big monster and refers to himself as her ' protector ' . xxmaj the lead actor xxmaj ryan xxmaj xxunk does a pretty
2 xxmaj the xxmaj scene xxmaj that xxmaj explains xxmaj all was adequate and managed to explain all of the questions and mysterious dialogue bits throughout the movie but we were just checking them off a list . ( " oh , okay , that 's why xxmaj brad had that happen and xxmaj jonathan says this and … ") \n\n xxmaj what laughs we made were from the stupidity of the plot the xxmaj scene xxmaj that xxmaj explains xxmaj all was adequate and managed to explain all of the questions and mysterious dialogue bits throughout the movie but we were just checking them off a list . ( " oh , okay , that 's why xxmaj brad had that happen and xxmaj jonathan says this and … ") \n\n xxmaj what laughs we made were from the stupidity of the plot than

TextDataLoaders.from_csv[source]

TextDataLoaders.from_csv(path, csv_fname='labels.csv', header='infer', delimiter=None, valid_pct=0.2, seed=None, text_col=0, label_col=1, label_delim=None, y_block=None, text_vocab=None, is_lm=False, valid_col=None, tok_tfm=None, seq_len=72, bs=64, val_bs=None, shuffle_train=True, device=None)

Create from csv file in path/csv_fname

Opens the csv file with header and delimiter, then pass all the other arguments to TextDataLoaders.from_df.

dls = TextDataLoaders.from_csv(path=path, csv_fname='texts.csv', text_col='text', label_col='label', valid_col='is_valid')
dls.show_batch(max_n=3)
text category
0 xxbos xxmaj raising xxmaj victor xxmaj vargas : a xxmaj review \n\n xxmaj you know , xxmaj raising xxmaj victor xxmaj vargas is like sticking your hands into a big , xxunk bowl of xxunk . xxmaj it 's warm and gooey , but you 're not sure if it feels right . xxmaj try as i might , no matter how warm and gooey xxmaj raising xxmaj victor xxmaj vargas became i was always aware that something did n't quite feel right . xxmaj victor xxmaj vargas suffers from a certain xxunk on the director 's part . xxmaj apparently , the director thought that the ethnic backdrop of a xxmaj latino family on the lower east side , and an xxunk storyline would make the film critic proof . xxmaj he was right , but it did n't fool me . xxmaj raising xxmaj victor xxmaj vargas is negative
1 xxbos xxup the xxup shop xxup around xxup the xxup corner is one of the xxunk and most feel - good romantic comedies ever made . xxmaj there 's just no getting around that , and it 's hard to actually put one 's feeling for this film into words . xxmaj it 's not one of those films that tries too hard , nor does it come up with the xxunk possible scenarios to get the two protagonists together in the end . xxmaj in fact , all its charm is xxunk , contained within the characters and the setting and the plot … which is highly believable to xxunk . xxmaj it 's easy to think that such a love story , as beautiful as any other ever told , * could * happen to you … a feeling you do n't often get from other romantic comedies positive
2 xxbos xxmaj now that xxmaj che(2008 ) has finished its relatively short xxmaj australian cinema run ( extremely limited xxunk screen in xxmaj xxunk , after xxunk ) , i can xxunk join both xxunk of " at xxmaj the xxmaj movies " in taking xxmaj steven xxmaj soderbergh to task . \n\n xxmaj it 's usually satisfying to watch a film director change his style / subject , but xxmaj soderbergh 's most recent stinker , xxmaj the xxmaj girlfriend xxmaj xxunk ) , was also missing a story , so narrative ( and editing ? ) seem to suddenly be xxmaj soderbergh 's main challenge . xxmaj strange , after xxunk years in the business . xxmaj he was probably never much good at narrative , just xxunk it well inside " edgy " projects . \n\n xxmaj none of this excuses him this present , almost diabolical negative