Core text modules

Contain the modules common between different architectures and the generic functions to get models

Language models

---

source

LinearDecoder

def LinearDecoder(
    n_out:int, # Number of output channels
    n_hid:int, # Number of features in encoder last layer output
    output_p:float=0.1, # Input dropout probability
    tie_encoder:torch.nn.modules.module.Module=None, # If module is supplied will tie decoder weight to `tie_encoder.weight`
    bias:bool=True, # If `False` the layer will not learn additive bias
):

To go on top of a RNNCore module and create a Language Model.

from fastai.text.models.awdlstm import *

enc = AWD_LSTM(100, 20, 10, 2)
x = torch.randint(0, 100, (10,5))
r = enc(x)

tst = LinearDecoder(100, 20, 0.1)
y = tst(r)
test_eq(y[1], r)
test_eq(y[2].shape, r.shape)
test_eq(y[0].shape, [10, 5, 100])

tst = LinearDecoder(100, 20, 0.1, tie_encoder=enc.encoder)
test_eq(tst.decoder.weight, enc.encoder.weight)

---

source

SequentialRNN

def SequentialRNN(
    *args
):

A sequential module that passes the reset call to its children.

class _TstMod(Module):
    def reset(self): print('reset')

tst = SequentialRNN(_TstMod(), _TstMod())
test_stdout(tst.reset, 'reset\nreset')

---

source

get_language_model

def get_language_model(
    arch, # Function or class that can generate a language model architecture
    vocab_sz:int, # Size of the vocabulary
    config:dict=None, # Model configuration dictionary
    drop_mult:float=1.0, # Multiplicative factor to scale all dropout probabilities in `config`
)->__main__.SequentialRNN: # Language model with `arch` encoder and linear decoder

Create a language model from arch and its config.

The default config used can be found in _model_meta[arch]['config_lm']. drop_mult is applied to all the probabilities of dropout in that config.

config = awd_lstm_lm_config.copy()
config.update({'n_hid':10, 'emb_sz':20})

tst = get_language_model(AWD_LSTM, 100, config=config)
x = torch.randint(0, 100, (10,5))
y = tst(x)
test_eq(y[0].shape, [10, 5, 100])
test_eq(y[1].shape, [10, 5, 20])
test_eq(y[2].shape, [10, 5, 20])
test_eq(tst[1].decoder.weight, tst[0].encoder.weight)

#test drop_mult
tst = get_language_model(AWD_LSTM, 100, config=config, drop_mult=0.5)
test_eq(tst[1].output_dp.p, config['output_p']*0.5)
for rnn in tst[0].rnns: test_eq(rnn.weight_p, config['weight_p']*0.5)
for dp in tst[0].hidden_dps: test_eq(dp.p, config['hidden_p']*0.5)
test_eq(tst[0].encoder_dp.embed_p, config['embed_p']*0.5)
test_eq(tst[0].input_dp.p, config['input_p']*0.5)

Classification models

---

source

SentenceEncoder

def SentenceEncoder(
    bptt:int, # Backpropagation through time
    module:torch.nn.modules.module.Module, # A module that can process up to [`bs`, `bptt`] tokens
    pad_idx:int=1, # Padding token id
    max_len:int=None, # Maximal output length
):

Create an encoder over module that can process a full sentence.

Warning

This module expects the inputs padded with most of the padding first, with the sequence beginning at a round multiple of bptt (and the rest of the padding at the end). Use pad_input_chunk to get your data in a suitable format.

mod = nn.Embedding(5, 10)
tst = SentenceEncoder(5, mod, pad_idx=0)
x = torch.randint(1, 5, (3, 15))
x[2,:5]=0
out,mask = tst(x)

test_eq(out[:1], mod(x)[:1])
test_eq(out[2,5:], mod(x)[2,5:])
test_eq(mask, x==0)

---

source

masked_concat_pool

def masked_concat_pool(
    output:torch.Tensor, # Output of sentence encoder
    mask:torch.Tensor, # Boolean mask as returned by sentence encoder
    bptt:int, # Backpropagation through time
)->torch.Tensor: # Concatenation of [last_hidden, max_pool, avg_pool]

Pool MultiBatchEncoder outputs into one vector [last_hidden, max_pool, avg_pool]

out = torch.randn(2,4,5)
mask = tensor([[True,True,False,False], [False,False,False,True]])
x = masked_concat_pool(out, mask, 2)

test_close(x[0,:5], out[0,-1])
test_close(x[1,:5], out[1,-2])
test_close(x[0,5:10], out[0,2:].max(dim=0)[0])
test_close(x[1,5:10], out[1,:3].max(dim=0)[0])
test_close(x[0,10:], out[0,2:].mean(dim=0))
test_close(x[1,10:], out[1,:3].mean(dim=0))

#Test the result is independent of padding by replacing the padded part by some random content
out1 = torch.randn(2,4,5)
out1[0,2:] = out[0,2:].clone()
out1[1,:3] = out[1,:3].clone()
x1 = masked_concat_pool(out1, mask, 2)
test_eq(x, x1)

---

source

PoolingLinearClassifier

def PoolingLinearClassifier(
    dims:list, # List of hidden sizes for MLP as `int`s
    ps:list, # List of dropout probabilities as `float`s
    bptt:int, # Backpropagation through time
    y_range:tuple=None, # Tuple of (low, high) output value bounds
):

Create a linear classifier with pooling

mod = nn.Embedding(5, 10)
tst = SentenceEncoder(5, mod, pad_idx=0)
x = torch.randint(1, 5, (3, 15))
x[2,:5]=0
out,mask = tst(x)

test_eq(out[:1], mod(x)[:1])
test_eq(out[2,5:], mod(x)[2,5:])
test_eq(mask, x==0)

---

source

get_text_classifier

def get_text_classifier(
    arch:Callable, # Function or class that can generate a language model architecture
    vocab_sz:int, # Size of the vocabulary
    n_class:int, # Number of classes
    seq_len:int=72, # Backpropagation through time
    config:dict=None, # Encoder configuration dictionary
    drop_mult:float=1.0, # Multiplicative factor to scale all dropout probabilities in `config`
    lin_ftrs:list=None, # List of hidden sizes for classifier head as `int`s
    ps:list=None, # List of dropout probabilities for classifier head as `float`s
    pad_idx:int=1, # Padding token id
    max_len:int=1440, # Maximal output length for `SentenceEncoder`
    y_range:tuple=None, # Tuple of (low, high) output value bounds
):

Create a text classifier from arch and its config, maybe pretrained

config = awd_lstm_clas_config.copy()
config.update({'n_hid':10, 'emb_sz':20})

tst = get_text_classifier(AWD_LSTM, 100, 3, config=config)
x = torch.randint(2, 100, (10,5))
y = tst(x)
test_eq(y[0].shape, [10, 3])
test_eq(y[1].shape, [10, 5, 20])
test_eq(y[2].shape, [10, 5, 20])

#test padding gives same results
tst.eval()
y = tst(x)
x1 = torch.cat([x, tensor([2,1,1,1,1,1,1,1,1,1])[:,None]], dim=1)
y1 = tst(x1)
test_close(y[0][1:],y1[0][1:])

#test drop_mult
tst = get_text_classifier(AWD_LSTM, 100, 3, config=config, drop_mult=0.5)
test_eq(tst[1].layers[1][1].p, 0.1)
test_eq(tst[1].layers[0][1].p, config['output_p']*0.5)
for rnn in tst[0].module.rnns: test_eq(rnn.weight_p, config['weight_p']*0.5)
for dp in tst[0].module.hidden_dps: test_eq(dp.p, config['hidden_p']*0.5)
test_eq(tst[0].module.encoder_dp.embed_p, config['embed_p']*0.5)
test_eq(tst[0].module.input_dp.p, config['input_p']*0.5)