In [ ]:

from fastai import *        # Quick access to most common functionality
from fastai.text import *   # Quick access to NLP functionality

Text example¶

An example of creating a language model and then transfering to a classifier.

In [ ]:

path = untar_data(URLs.IMDB_SAMPLE)
path

Open and view the independent and dependent variables:

In [ ]:

df = pd.read_csv(path/'train.csv', header=None)
df.head()

In [ ]:

classes = read_classes(path/'classes.txt')
classes[0], classes[1]

Create a DataBunch for each of the language model and the classifier:

In [ ]:

data_lm = TextLMDataBunch.from_csv(path)
data_clas = TextClasDataBunch.from_csv(path, vocab=data_lm.train_ds.vocab)

fast.ai has a pre-trained English model available that we can download.

In [ ]:

URLs.download_wt103_model()

We'll fine-tune the language model:

In [ ]:

learn = RNNLearner.language_model(data_lm, pretrained_fnames=['lstm_wt103', 'itos_wt103'])
learn.unfreeze()
learn.fit(2, slice(1e-4,1e-2))

In [ ]:

learn.loss_func

Save our language model's encoder:

In [ ]:

learn.save_encoder('enc')

Fine tune it to create a classifier:

In [ ]:

learn = RNNLearner.classifier(data_clas)
learn.load_encoder('enc')
learn.fit(3, 1e-3)