tape/models/modeling_bert.py

# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors and The HuggingFace Inc. team.
# Copyright (c) 2018, NVIDIA CORPORATION.  All rights reserved.
# Modified by Roshan Rao
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""PyTorch BERT model. """

from __future__ import absolute_import, division, print_function, unicode_literals

import logging
import math

import torch
from torch import nn
from torch.utils.checkpoint import checkpoint

from .modeling_utils import ProteinConfig
from .modeling_utils import ProteinModel
from .modeling_utils import prune_linear_layer
from .modeling_utils import get_activation_fn
from .modeling_utils import LayerNorm
from .modeling_utils import MLMHead
from .modeling_utils import ValuePredictionHead
from .modeling_utils import SequenceClassificationHead
from .modeling_utils import SequenceToSequenceClassificationHead
from .modeling_utils import PairwiseContactPredictionHead
from ..registry import registry

logger = logging.getLogger(__name__)

URL_PREFIX = "https://s3.amazonaws.com/proteindata/pytorch-models/"
BERT_PRETRAINED_MODEL_ARCHIVE_MAP = {
    'bert-base': URL_PREFIX + "bert-base-pytorch_model.bin",
}
BERT_PRETRAINED_CONFIG_ARCHIVE_MAP = {
    'bert-base': URL_PREFIX + "bert-base-config.json"
}


class ProteinBertConfig(ProteinConfig):
    r"""
        :class:`~pytorch_transformers.ProteinBertConfig` is the configuration class to store the
        configuration of a `ProteinBertModel`.


        Arguments:
            vocab_size_or_config_json_file: Vocabulary size of `inputs_ids` in
                `ProteinBertModel`.
            hidden_size: Size of the encoder layers and the pooler layer.
            num_hidden_layers: Number of hidden layers in the ProteinBert encoder.
            num_attention_heads: Number of attention heads for each attention layer in
                the ProteinBert encoder.
            intermediate_size: The size of the "intermediate" (i.e., feed-forward)
                layer in the ProteinBert encoder.
            hidden_act: The non-linear activation function (function or string) in the
                encoder and pooler. If string, "gelu", "relu" and "swish" are supported.
            hidden_dropout_prob: The dropout probabilitiy for all fully connected
                layers in the embeddings, encoder, and pooler.
            attention_probs_dropout_prob: The dropout ratio for the attention
                probabilities.
            max_position_embeddings: The maximum sequence length that this model might
                ever be used with. Typically set this to something large just in case
                (e.g., 512 or 1024 or 2048).
            type_vocab_size: The vocabulary size of the `token_type_ids` passed into
                `ProteinBertModel`.
            initializer_range: The sttdev of the truncated_normal_initializer for
                initializing all weight matrices.
            layer_norm_eps: The epsilon used by LayerNorm.
    """
    pretrained_config_archive_map = BERT_PRETRAINED_CONFIG_ARCHIVE_MAP

    def __init__(self,
                 vocab_size: int = 30,
                 hidden_size: int = 768,
                 num_hidden_layers: int = 12,
                 num_attention_heads: int = 12,
                 intermediate_size: int = 3072,
                 hidden_act: str = "gelu",
                 hidden_dropout_prob: float = 0.1,
                 attention_probs_dropout_prob: float = 0.1,
                 max_position_embeddings: int = 8096,
                 type_vocab_size: int = 2,
                 initializer_range: float = 0.02,
                 layer_norm_eps: float = 1e-12,
                 temporal_pooling: str = 'attention',
                 freeze_embedding: bool = False,
                 **kwargs):
        super().__init__(**kwargs)
        self.vocab_size = vocab_size
        self.hidden_size = hidden_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.hidden_act = hidden_act
        self.intermediate_size = intermediate_size
        self.hidden_dropout_prob = hidden_dropout_prob
        self.attention_probs_dropout_prob = attention_probs_dropout_prob
        self.max_position_embeddings = max_position_embeddings
        self.type_vocab_size = type_vocab_size
        self.initializer_range = initializer_range
        self.layer_norm_eps = layer_norm_eps
        self.temporal_pooling = temporal_pooling
        self.freeze_embedding = freeze_embedding


class ProteinBertEmbeddings(nn.Module):
    """Construct the embeddings from word, position and token_type embeddings.
    """
    def __init__(self, config):
        super().__init__()
        self.word_embeddings = nn.Embedding(
            config.vocab_size, config.hidden_size, padding_idx=0)
        self.position_embeddings = nn.Embedding(
            config.max_position_embeddings, config.hidden_size)
        self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)

        # self.LayerNorm is not snake-cased to stick with TensorFlow model variable name and be
        # able to load any TensorFlow checkpoint file
        self.LayerNorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, input_ids, token_type_ids=None, position_ids=None):
        seq_length = input_ids.size(1)
        if position_ids is None:
            position_ids = torch.arange(seq_length, dtype=torch.long, device=input_ids.device)
            position_ids = position_ids.unsqueeze(0).expand_as(input_ids)
        if token_type_ids is None:
            token_type_ids = torch.zeros_like(input_ids)

        words_embeddings = self.word_embeddings(input_ids)
        position_embeddings = self.position_embeddings(position_ids)
        token_type_embeddings = self.token_type_embeddings(token_type_ids)

        embeddings = words_embeddings + position_embeddings + token_type_embeddings
        embeddings = self.LayerNorm(embeddings)
        embeddings = self.dropout(embeddings)
        return embeddings


class ProteinBertSelfAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        if config.hidden_size % config.num_attention_heads != 0:
            raise ValueError(
                "The hidden size (%d) is not a multiple of the number of attention "
                "heads (%d)" % (config.hidden_size, config.num_attention_heads))
        self.output_attentions = config.output_attentions

        self.num_attention_heads = config.num_attention_heads
        self.attention_head_size = int(config.hidden_size / config.num_attention_heads)
        self.all_head_size = self.num_attention_heads * self.attention_head_size

        self.query = nn.Linear(config.hidden_size, self.all_head_size)
        self.key = nn.Linear(config.hidden_size, self.all_head_size)
        self.value = nn.Linear(config.hidden_size, self.all_head_size)

        self.dropout = nn.Dropout(config.attention_probs_dropout_prob)

    def transpose_for_scores(self, x):
        new_x_shape = x.size()[:-1] + (self.num_attention_heads, self.attention_head_size)
        x = x.view(*new_x_shape)
        return x.permute(0, 2, 1, 3)

    def forward(self, hidden_states, attention_mask):
        mixed_query_layer = self.query(hidden_states)
        mixed_key_layer = self.key(hidden_states)
        mixed_value_layer = self.value(hidden_states)

        query_layer = self.transpose_for_scores(mixed_query_layer)
        key_layer = self.transpose_for_scores(mixed_key_layer)
        value_layer = self.transpose_for_scores(mixed_value_layer)

        # Take the dot product between "query" and "key" to get the raw attention scores.
        attention_scores = torch.matmul(query_layer, key_layer.transpose(-1, -2))
        attention_scores = attention_scores / math.sqrt(self.attention_head_size)
        # Apply the attention mask is (precomputed for all layers in
        # ProteinBertModel forward() function)
        attention_scores = attention_scores + attention_mask

        # Normalize the attention scores to probabilities.
        attention_probs = nn.Softmax(dim=-1)(attention_scores)

        # This is actually dropping out entire tokens to attend to, which might
        # seem a bit unusual, but is taken from the original ProteinBert paper.
        attention_probs = self.dropout(attention_probs)

        context_layer = torch.matmul(attention_probs, value_layer)

        context_layer = context_layer.permute(0, 2, 1, 3).contiguous()
        new_context_layer_shape = context_layer.size()[:-2] + (self.all_head_size,)
        context_layer = context_layer.view(*new_context_layer_shape)

        outputs = (context_layer, attention_probs) \
            if self.output_attentions else (context_layer,)
        return outputs


class ProteinBertSelfOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.LayerNorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class ProteinBertAttention(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.self = ProteinBertSelfAttention(config)
        self.output = ProteinBertSelfOutput(config)

    def prune_heads(self, heads):
        if len(heads) == 0:
            return
        mask = torch.ones(self.self.num_attention_heads, self.self.attention_head_size)
        for head in heads:
            mask[head] = 0
        mask = mask.view(-1).contiguous().eq(1)
        index = torch.arange(len(mask))[mask].long()
        # Prune linear layers
        self.self.query = prune_linear_layer(self.self.query, index)
        self.self.key = prune_linear_layer(self.self.key, index)
        self.self.value = prune_linear_layer(self.self.value, index)
        self.output.dense = prune_linear_layer(self.output.dense, index, dim=1)
        # Update hyper params
        self.self.num_attention_heads = self.self.num_attention_heads - len(heads)
        self.self.all_head_size = self.self.attention_head_size * self.self.num_attention_heads

    def forward(self, input_tensor, attention_mask):
        self_outputs = self.self(input_tensor, attention_mask)
        attention_output = self.output(self_outputs[0], input_tensor)
        outputs = (attention_output,) + self_outputs[1:]  # add attentions if we output them
        return outputs


class ProteinBertIntermediate(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.intermediate_size)
        if isinstance(config.hidden_act, str):
            self.intermediate_act_fn = get_activation_fn(config.hidden_act)
        else:
            self.intermediate_act_fn = config.hidden_act

    def forward(self, hidden_states):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.intermediate_act_fn(hidden_states)
        return hidden_states


class ProteinBertOutput(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.intermediate_size, config.hidden_size)
        self.LayerNorm = LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
        self.dropout = nn.Dropout(config.hidden_dropout_prob)

    def forward(self, hidden_states, input_tensor):
        hidden_states = self.dense(hidden_states)
        hidden_states = self.dropout(hidden_states)
        hidden_states = self.LayerNorm(hidden_states + input_tensor)
        return hidden_states


class ProteinBertLayer(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.attention = ProteinBertAttention(config)
        self.intermediate = ProteinBertIntermediate(config)
        self.output = ProteinBertOutput(config)

    def forward(self, hidden_states, attention_mask):
        attention_outputs = self.attention(hidden_states, attention_mask)
        attention_output = attention_outputs[0]
        intermediate_output = self.intermediate(attention_output)
        layer_output = self.output(intermediate_output, attention_output)
        outputs = (layer_output,) + attention_outputs[1:]  # add attentions if we output them
        return outputs


class ProteinBertEncoder(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.output_attentions = config.output_attentions
        self.output_hidden_states = config.output_hidden_states
        self.layer = nn.ModuleList(
            [ProteinBertLayer(config) for _ in range(config.num_hidden_layers)])

    def run_function(self, start, chunk_size):
        def custom_forward(hidden_states, attention_mask):
            all_hidden_states = ()
            all_attentions = ()
            chunk_slice = slice(start, start + chunk_size)
            for layer in self.layer[chunk_slice]:
                if self.output_hidden_states:
                    all_hidden_states = all_hidden_states + (hidden_states,)
                layer_outputs = layer(hidden_states, attention_mask)
                hidden_states = layer_outputs[0]

                if self.output_attentions:
                    all_attentions = all_attentions + (layer_outputs[1],)

            if self.output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)
            outputs = (hidden_states,)
            if self.output_hidden_states:
                outputs = outputs + (all_hidden_states,)
            if self.output_attentions:
                outputs = outputs + (all_attentions,)
            return outputs

        return custom_forward

    def forward(self, hidden_states, attention_mask, chunks=None):
        all_hidden_states = ()
        all_attentions = ()

        if chunks is not None:
            assert isinstance(chunks, int)
            chunk_size = (len(self.layer) + chunks - 1) // chunks
            for start in range(0, len(self.layer), chunk_size):
                outputs = checkpoint(self.run_function(start, chunk_size),
                                     hidden_states, attention_mask)
                if self.output_hidden_states:
                    all_hidden_states = all_hidden_states + outputs[1]
                if self.output_attentions:
                    all_attentions = all_attentions + outputs[-1]
                hidden_states = outputs[0]
        else:
            for i, layer_module in enumerate(self.layer):
                if self.output_hidden_states:
                    all_hidden_states = all_hidden_states + (hidden_states,)

                layer_outputs = layer_module(hidden_states, attention_mask)
                hidden_states = layer_outputs[0]

                if self.output_attentions:
                    all_attentions = all_attentions + (layer_outputs[1],)

            # Add last layer
            if self.output_hidden_states:
                all_hidden_states = all_hidden_states + (hidden_states,)

            outputs = (hidden_states,)
            if self.output_hidden_states:
                outputs = outputs + (all_hidden_states,)
            if self.output_attentions:
                outputs = outputs + (all_attentions,)
        return outputs  # outputs, (hidden states), (attentions)


class ProteinBertPooler(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
        self.activation = nn.Tanh()
        self.temporal_pooling = config.temporal_pooling
        self._la_w1 = nn.Conv1d(config.hidden_size, int(config.hidden_size/2), 5, padding=2)
        self._la_w2 = nn.Conv1d(config.hidden_size, int(config.hidden_size/2), 5, padding=2)
        self._la_mlp = nn.Linear(config.hidden_size, config.hidden_size)

    def forward(self, hidden_states):
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        if self.temporal_pooling == 'mean':
            return hidden_states.mean(dim=1)
        if self.temporal_pooling == 'max':
            return hidden_states.max(dim=1)
        if self.temporal_pooling == 'concat':
            _temp = hidden_states.reshape(hidden_states.shape[0], -1)
            return torch.nn.functional.pad(_temp, (0, 2048 - _temp.shape[1]))
        if self.temporal_pooling == 'topmax':
            val, _ = torch.topk(hidden_states, k=5, dim=1)
            return val.mean(dim=1)
        if self.temporal_pooling == 'light_attention':
            _temp = hidden_states.permute(0,2,1)
            a = self._la_w1(_temp).softmax(dim=-1)
            v = self._la_w2(_temp)
            v_max = v.max(dim=-1).values
            v_sum = (a * v).sum(dim=-1)
            return self._la_mlp(torch.cat([v_max, v_sum], dim=1))

        first_token_tensor = hidden_states[:, 0]
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output


class ProteinBertAbstractModel(ProteinModel):
    """ An abstract class to handle weights initialization and
        a simple interface for dowloading and loading pretrained models.
    """
    config_class = ProteinBertConfig
    pretrained_model_archive_map = BERT_PRETRAINED_MODEL_ARCHIVE_MAP
    base_model_prefix = "bert"

    def _init_weights(self, module):
        """ Initialize the weights """
        if isinstance(module, (nn.Linear, nn.Embedding)):
            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
        elif isinstance(module, LayerNorm):
            module.bias.data.zero_()
            module.weight.data.fill_(1.0)
        if isinstance(module, nn.Linear) and module.bias is not None:
            module.bias.data.zero_()


@registry.register_task_model('embed', 'transformer')
class ProteinBertModel(ProteinBertAbstractModel):

    def __init__(self, config):
        super().__init__(config)

        self.embeddings = ProteinBertEmbeddings(config)
        self.encoder = ProteinBertEncoder(config)
        self.pooler = ProteinBertPooler(config)

        self.init_weights()

    def _resize_token_embeddings(self, new_num_tokens):
        old_embeddings = self.embeddings.word_embeddings
        new_embeddings = self._get_resized_embeddings(old_embeddings, new_num_tokens)
        self.embeddings.word_embeddings = new_embeddings
        return self.embeddings.word_embeddings

    def _prune_heads(self, heads_to_prune):
        """ Prunes heads of the model.
            heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
            See base class ProteinModel
        """
        for layer, heads in heads_to_prune.items():
            self.encoder.layer[layer].attention.prune_heads(heads)

    def forward(self,
                input_ids,
                input_mask=None):
        if input_mask is None:
            input_mask = torch.ones_like(input_ids)

        # We create a 3D attention mask from a 2D tensor mask.
        # Sizes are [batch_size, 1, 1, to_seq_length]
        # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
        # this attention mask is more simple than the triangular masking of causal attention
        # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
        extended_attention_mask = input_mask.unsqueeze(1).unsqueeze(2)

        # Since input_mask is 1.0 for positions we want to attend and 0.0 for
        # masked positions, this operation will create a tensor which is 0.0 for
        # positions we want to attend and -10000.0 for masked positions.
        # Since we are adding it to the raw scores before the softmax, this is
        # effectively the same as removing these entirely.
        extended_attention_mask = extended_attention_mask.to(
            dtype=torch.float32)  # fp16 compatibility
        extended_attention_mask = (1.0 - extended_attention_mask) * -10000.0

        embedding_output = self.embeddings(input_ids)
        encoder_outputs = self.encoder(embedding_output,
                                       extended_attention_mask,
                                       chunks=None)
        sequence_output = encoder_outputs[0]
        pooled_output = self.pooler(sequence_output)

        # add hidden_states and attentions if they are here
        outputs = (sequence_output, pooled_output,) + encoder_outputs[1:]
        return outputs  # sequence_output, pooled_output, (hidden_states), (attentions)


@registry.register_task_model('masked_language_modeling', 'transformer')
class ProteinBertForMaskedLM(ProteinBertAbstractModel):

    def __init__(self, config):
        super().__init__(config)

        self.bert = ProteinBertModel(config)
        self.mlm = MLMHead(
            config.hidden_size, config.vocab_size, config.hidden_act, config.layer_norm_eps,
            ignore_index=-1)

        self.init_weights()
        self.tie_weights()

    def tie_weights(self):
        """ Make sure we are sharing the input and output embeddings.
            Export to TorchScript can't handle parameter sharing so we are cloning them instead.
        """
        self._tie_or_clone_weights(self.mlm.decoder,
                                   self.bert.embeddings.word_embeddings)

    def forward(self,
                input_ids,
                input_mask=None,
                targets=None):

        outputs = self.bert(input_ids, input_mask=input_mask)

        sequence_output, pooled_output = outputs[:2]
        # add hidden states and attention if they are here
        outputs = self.mlm(sequence_output, targets) + outputs[:2]
        # (loss), prediction_scores, (hidden_states), (attentions)
        return outputs


@registry.register_task_model('fluorescence', 'transformer')
@registry.register_task_model('stability', 'transformer')
class ProteinBertForValuePrediction(ProteinBertAbstractModel):

    def __init__(self, config):
        super().__init__(config)

        self.bert = ProteinBertModel(config)
        self.predict = ValuePredictionHead(config.hidden_size)
        self.freeze_embedding = config.freeze_embedding
        self.init_weights()

    def forward(self, input_ids, input_mask=None, targets=None):
        if self.freeze_embedding:
            self.bert.train(False)
        outputs = self.bert(input_ids, input_mask=input_mask)

        sequence_output, pooled_output = outputs[:2]
        outputs = self.predict(pooled_output, targets) + outputs[2:]
        # (loss), prediction_scores, (hidden_states), (attentions)
        return outputs


@registry.register_task_model('remote_homology', 'transformer')
class ProteinBertForSequenceClassification(ProteinBertAbstractModel):

    def __init__(self, config):
        super().__init__(config)

        self.bert = ProteinBertModel(config)
        self.classify = SequenceClassificationHead(
            config.hidden_size, config.num_labels)
        self.freeze_embedding = config.freeze_embedding
        self.init_weights()

    def forward(self, input_ids, input_mask=None, targets=None):
        if self.freeze_embedding:
            self.bert.train(False)
        outputs = self.bert(input_ids, input_mask=input_mask)

        sequence_output, pooled_output = outputs[:2]

        outputs = self.classify(pooled_output, targets) + outputs[2:]
        # (loss), prediction_scores, (hidden_states), (attentions)
        return outputs


@registry.register_task_model('secondary_structure', 'transformer')
class ProteinBertForSequenceToSequenceClassification(ProteinBertAbstractModel):

    def __init__(self, config):
        super().__init__(config)

        self.bert = ProteinBertModel(config)
        self.classify = SequenceToSequenceClassificationHead(
            config.hidden_size, config.num_labels, ignore_index=-1)

        self.init_weights()

    def forward(self, input_ids, input_mask=None, targets=None):

        outputs = self.bert(input_ids, input_mask=input_mask)

        sequence_output, pooled_output = outputs[:2]
        outputs = self.classify(sequence_output, targets) + outputs[2:]
        # (loss), prediction_scores, (hidden_states), (attentions)
        return outputs


@registry.register_task_model('contact_prediction', 'transformer')
class ProteinBertForContactPrediction(ProteinBertAbstractModel):

    def __init__(self, config):
        super().__init__(config)

        self.bert = ProteinBertModel(config)
        self.predict = PairwiseContactPredictionHead(config.hidden_size, ignore_index=-1)

        self.init_weights()

    def forward(self, input_ids, protein_length, input_mask=None, targets=None):

        outputs = self.bert(input_ids, input_mask=input_mask)

        sequence_output, pooled_output = outputs[:2]
        outputs = self.predict(sequence_output, protein_length, targets) + outputs[2:]
        # (loss), prediction_scores, (hidden_states), (attentions)
        return outputs