#!/usr/bin/env python3
# -*- coding: utf-8 -*-
#
# MIT License
#
# Copyright (c) 2018 Kim Seonghyeon
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
#
# ------------------------------------------------------------------------------
#
# Copyright (C) IBM Corporation 2018
#
# 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.
"""
read_unit.py: Implementation of the ``ReadUnit`` for the MAC network. Cf https://arxiv.org/abs/1803.03067 for the \
reference paper.
"""
__author__ = "Vincent Marois"
import torch
from torch.nn import Module
from miprometheus.models.mac.utils_mac import linear
[docs]class ReadUnit(Module):
"""
Implementation of the ``ReadUnit`` of the MAC network.
"""
[docs] def __init__(self, dim):
"""
Constructor for the ``ReadUnit``.
:param dim: global 'd' hidden dimension
:type dim: int
"""
# call base constructor
super(ReadUnit, self).__init__()
# define linear layer for the projection of the previous memory state
self.mem_proj_layer = linear(dim, dim, bias=True)
# linear layer to define I'(i,h,w) elements (r2 equation)
self.concat_layer = linear(2 * dim, dim, bias=True)
# linear layer to compute attention weights
self.attn = linear(dim, 1, bias=True)
[docs] def forward(self, memory_states, knowledge_base, ctrl_states, kb_proj):
"""
Forward pass of the ``ReadUnit``. Assuming 1 scalar attention weight per \
knowledge base elements.
:param memory_states: list of all previous memory states, each of shape [batch_size x mem_dim]
:type memory_states: torch.tensor
:param knowledge_base: image representation (output of CNN), shape [batch_size x nb_kernels x (feat_H * feat_W)]
:type knowledge_base: torch.tensor
:param ctrl_states: All previous control state, each of shape [batch_size x ctrl_dim].
:type ctrl_states: list
:return: current read vector, shape [batch_size x read_dim]
"""
# assume mem_dim = ctrl_dim = nb_kernels = dim
# retrieve the last memory & control state
memory_state = memory_states[-1]
ctrl_state = ctrl_states[-1]
# pass memory state through linear layer
memory_state = self.mem_proj_layer(memory_state).unsqueeze(2)
# memory_state: [batch_size x dim x 1]
# compute I(i,h,w) elements (r1 equation)
# [batch_size x dim x 1] * [batch_size x dim x (H*W)] -> [batch_size x dim x (H*W)]
I_elements = memory_state * kb_proj
# compute I' elements (r2 equation)
concat = self.concat_layer(
torch.cat([I_elements, knowledge_base],
dim=1).permute(0, 2, 1)) # [batch_size x (H*W) x dim]
# compute attention weights
rai = self.attn(concat * ctrl_state.unsqueeze(1)
).squeeze(2) # [batch_size x (H*W)]
# This is for the time plot
self.rvi = torch.nn.functional.softmax(rai, 1).unsqueeze(1) # [batch_size x 1 x (H*W)]
# apply attn weights on knowledge base elements & sum on (H*W)
read_vector = (self.rvi * knowledge_base).sum(2) # [batch_size x dim]
return read_vector