TransCNN.py

import torch.nn as nn
import torch
from torch.nn.init import *
import torch.nn.functional as F
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import math, copy, time


def clones(module, N):
    "Produce N identical layers."
    return nn.ModuleList([copy.deepcopy(module) for _ in range(N)])


class Encoder(nn.Module):
    "Core encoder is a stack of N layers"

    def __init__(self, layer, N):
        super(Encoder, self).__init__()
        self.layers = clones(layer, N)
        self.norm = LayerNorm(layer.size)

    def forward(self, x, mask=None):
        "Pass the input (and mask) through each layer in turn."
        for layer in self.layers:
            x = layer(x, mask)
        return self.norm(x)


class PositionwiseFeedForward(nn.Module):
    def __init__(self, d_model, d_ff, dropout=0.1):
        super(PositionwiseFeedForward, self).__init__()
        self.w_1 = nn.Linear(d_model, d_ff)
        self.w_2 = nn.Linear(d_ff, d_model)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        return self.w_2(self.dropout(F.relu(self.w_1(x))))


class LayerNorm(nn.Module):
    "Construct a layernorm module (See citation for details)."

    def __init__(self, features, eps=1e-6):
        super(LayerNorm, self).__init__()
        self.a_2 = nn.Parameter(torch.ones(features))
        self.b_2 = nn.Parameter(torch.zeros(features))
        self.eps = eps

    def forward(self, x):
        mean = x.mean(-1, keepdim=True)
        std = x.std(-1, keepdim=True)
        return self.a_2 * (x - mean) / (std + self.eps) + self.b_2


class SublayerConnection(nn.Module):
    """
    A residual connection followed by a layer norm.
    Note for code simplicity the norm is first as opposed to last.
    """

    def __init__(self, size, dropout):
        super(SublayerConnection, self).__init__()
        self.norm = LayerNorm(size)
        self.dropout = nn.Dropout(dropout)

    def forward(self, x, sublayer):
        "Apply residual connection to any sublayer with the same size."
        return x + self.dropout(sublayer(self.norm(x)))


def attention(query, key, value, mask=None, dropout=None):
    "Compute 'Scaled Dot Product Attention'"
    d_k = query.size(-1)
    scores = torch.matmul(query, key.transpose(-2, -1)) \
             / math.sqrt(d_k)
    if mask is not None:
        scores = scores.masked_fill(mask == 0, -1e9)
    p_attn = F.softmax(scores, dim=-1)
    if dropout is not None:
        p_attn = dropout(p_attn)
    return torch.matmul(p_attn, value), p_attn


class HAR_CNN(nn.Module):
    "Implements CNN equation."
    def __init__(self, d_model, d_ff, filters, dropout=0.1):
        super(HAR_CNN, self).__init__()
        self.kernel_num = int(d_ff)
        self.filter_sizes = filters
        self.dropout = nn.Dropout(dropout)
        self.bn = nn.BatchNorm1d(d_model)
        self.encoders = []
        for i, filter_size in enumerate(self.filter_sizes):
            enc_attr_name = "encoder_%d" % i
            self.__setattr__(enc_attr_name,
                             torch.nn.Conv1d(
                                 in_channels=d_model,
                                 out_channels=self.kernel_num,
                                 kernel_size=filter_size,
                                 padding=int((filter_size-1)/2))
                             )
            self.encoders.append(self.__getattr__(enc_attr_name))

    def forward(self, x):
        enc_outs = []
        for encoder in self.encoders:
            f_map = encoder(x.transpose(-1, -2))
            enc_ = f_map
            #enc_ = F.relu(f_map)
            #k_h = enc_.size()[-1]
            #enc_ = F.max_pool1d(enc_, kernel_size=k_h)
            #enc_ = enc_.squeeze(dim=-1)
            enc_ = F.relu(self.dropout(self.bn(enc_)))
            enc_outs.append(enc_.unsqueeze(dim=1))
        re = torch.div(torch.sum(torch.cat(enc_outs, 1), dim=1), 3)
        encoding = re
        #encoding = self.dropout(torch.cat(enc_outs, 1))
        #q_re = F.relu(encoding)
        return encoding.transpose(-1, -2)

class EncoderLayer(nn.Module):
    "Encoder is made up of self-attn and feed forward (defined below)"

    def __init__(self, size, self_attn, feed_forward, dropout):
        super(EncoderLayer, self).__init__()
        self.self_attn = self_attn
        self.feed_forward = feed_forward
        self.sublayer = clones(SublayerConnection(size, dropout), 2)
        self.size = size

    def forward(self, x, mask=None):
        "Follow Figure 1 (left) for connections."
        x = self.sublayer[0](x, lambda x: self.self_attn(x, x, x, mask))
        return self.sublayer[1](x, self.feed_forward)


def attention_with_pos(query, key, value, pos_k, pos_v, mask=None, dropout=None):
    "Compute 'Scaled Dot Product Attention'"
    d_k = query.size(-1)
    scores = torch.matmul(query, key.transpose(-2, -1)) \
             / math.sqrt(d_k)
    if mask is not None:
        scores = scores.masked_fill(mask == 0, -1e9)
    p_attn = F.softmax(scores, dim=-1)
    if dropout is not None:
        p_attn = dropout(p_attn)
    return torch.matmul(p_attn, value), p_attn


class MultiHeadedAttention(nn.Module):
    def __init__(self, h, d_model, dropout=0.1):
        "Take in model size and number of heads."
        super(MultiHeadedAttention, self).__init__()
        assert d_model % h == 0
        # We assume d_v always equals d_k
        self.d_k = d_model // h
        self.h = h
        self.linears = clones(nn.Linear(d_model, d_model), 4)
        self.attn = None
        self.dropout = nn.Dropout(p=dropout)
        #relative positional encoding
        #self.k = 20
        #self.pos_k = torch.zeros([self.k, d_model], dtype=torch.float, requires_grad=True)
        #self.pos_v = torch.zeros([self.k, d_model], dtype=torch.float, requires_grad=True)
        #nn.init.xavier_uniform_(self.pos_k, gain=1)
        #nn.init.xavier_uniform_(self.pos_v, gain=1)

    def get_rel_pos(self, x):
        return max(self.k*-1, min(self.k, x))

    def forward(self, query, key, value, mask=None):
        "Implements Figure 2"
        if mask is not None:
            # Same mask applied to all h heads.
            mask = mask.unsqueeze(1)
        nbatches = query.size(0)

        # 1) Do all the linear projections in batch from d_model => h x d_k
        query, key, value = \
            [l(x).view(nbatches, -1, self.h, self.d_k).transpose(1, 2)
             for l, x in zip(self.linears, (query, key, value))]

        # 2) Apply attention on all the projected vectors in batch.
        x, self.attn = attention(query, key, value, mask=mask,
                                 dropout=self.dropout)

        # 3) "Concat" using a view and apply a final linear.
        x = x.transpose(1, 2).contiguous() \
            .view(nbatches, -1, self.h * self.d_k)
        return self.linears[-1](x)

def normal_pdf(pos, mu, sigma):
    a = pos - mu
    log_p = -1*torch.mul(a, a)/(2*sigma) - torch.log(sigma)/2
    return F.softmax(log_p, dim=1)


def get_pe(d_model, max_len=5000):
    # Compute the positional encodings once in log space.
    pe = torch.zeros(max_len, d_model)
    position = torch.arange(0, max_len).unsqueeze(1)
    div_term = torch.exp(torch.arange(0, d_model, 2) *
                         -(math.log(10000.0) / d_model))
    pe[:, 0::2] = torch.sin(position * div_term)
    pe[:, 1::2] = torch.cos(position * div_term)
    #pe.requires_grad = False
    return pe


class Gaussian_Position(nn.Module):
    def __init__(self, d_model, total_size, K=10):
        super(Gaussian_Position, self).__init__()
        #self.embedding = get_pe(d_model, K).to('cuda')
        #self.register_buffer('pe', self.embedding)
        self.embedding = nn.Parameter(torch.zeros([K, d_model], dtype=torch.float), requires_grad=True)
        nn.init.xavier_uniform_(self.embedding, gain=1)
        self.positions = torch.tensor([i for i in range(total_size)], requires_grad=False).unsqueeze(1).repeat(1, K).to('cuda')
        s = 0.0
        interval = total_size / K
        mu = []
        for _ in range(K):
            mu.append(nn.Parameter(torch.tensor(s, dtype=torch.float), requires_grad=True))
            s = s + interval
        self.mu = nn.Parameter(torch.tensor(mu, dtype=torch.float).unsqueeze(0), requires_grad=True)
        self.sigma = nn.Parameter(torch.tensor([torch.tensor([50.0], dtype=torch.float, requires_grad=True) for _ in range(K)]).unsqueeze(0))

    def forward(self, x):
        M = normal_pdf(self.positions, self.mu, self.sigma)
        pos_enc = torch.matmul(M, self.embedding)
        #print(M)
        return x + pos_enc.unsqueeze(0).repeat(x.size(0), 1, 1)


class PositionalEncoding(nn.Module):
    "Implement the PE function."

    def __init__(self, d_model, dropout, max_len=5000):
        super(PositionalEncoding, self).__init__()
        self.dropout = nn.Dropout(p=dropout)

        # Compute the positional encodings once in log space.
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2) *
                             -(math.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe.requires_grad = False
        #pe = pe.unsqueeze(0)
        self.register_buffer('pe', pe)

    def forward(self, x):
        #x = x + self.pe[:x.size(0), :]#Variable(self.pe[:x.size(0), :], requires_grad=False)
        x = x + self.pe[:x.size(1), :].unsqueeze(0).repeat(x.size(0), 1, 1) ## modified by Bing to adapt to batch
        return self.dropout(x)

class HARTransformer(nn.Module):

    def __init__(self, hidden_dim, N, H, total_size, filters=[1, 3, 5]):
        super(HARTransformer, self).__init__()
        #self.pos_encoding = get_pe(hidden_dim)
        self.model = Encoder(
            EncoderLayer(hidden_dim, MultiHeadedAttention(H, hidden_dim),
                         HAR_CNN(hidden_dim, hidden_dim, filters)
                         , 0.1),
            N
        )

    def forward(self, x, mask=None):
        #x = self.pos_encoding(x)
        return self.model(x, mask)