arunmallya

# Since convolution takes the most time, only do it on images with
# mask = 1. Note that masks.data.nonzero() is of size (N, 1).
# As a result, when expanding to 4 dims, we need to unsqueeze it twice.
selected_idx = Variable(masks.data.nonzero().unsqueeze(2).unsqueeze(3).repeat(
    1, images.size(1), images.size(2), images.size(3)))
selected_images = torch.gather(images, 0, selected_idx)

# Get image features from CNN and linear layer rnn_emb.
some_im_feats = self.rnn_emb(self.cnn(selected_images).squeeze())

arunmallya

import torch
import torch.nn as nn
from torch.autograd import Variable

class SimpleModel(nn.Module):
	def __init__(self):
		super(SimpleModel, self).__init__()
		self.net = nn.Linear(10, 2)

	def forward(self, inputs):

arunmallya

import torch.nn as nn

def myModule(nn.Module):
  def __init__(self):
    # Init stuff here
    self.X = nn.Sequential(
               nn.Linear(num_input_genes, num_tfs),
               nn.ReLU(),
               nn.BatchNorm1d(num_tfs)
             )

arunmallya

bool ReadCSVToDatum(const string& filename, Datum* datum) {
  // read in the CSV file into a vector
  fstream file(filename.c_str(), ios::in);
  vector<vector<int> > label;
  std::string line;
  while (std::getline(file, line)) {
    // replace commas with spaces
    for (int i = 0; i < line.length(); i++) {
      if (line[i] == ',')
        line[i] = ' ';

arunmallya

from __future__ import print_function
import torch

A = torch.rand(4)
idx = torch.LongTensor([0, 3])

An = A.numpy()
idxn = idx.numpy()

# Numpy indexing.

arunmallya

from __future__ import print_function
import torch

A = torch.rand(4, 4)
An = A.numpy()

idx = torch.ByteTensor([1, 0, 0, 1])
idxn = [True, False, False, True]

# Numpy indexing.

arunmallya

DEFAULT_THRESHOLD = 5e-3

class Binarizer(torch.autograd.Function):
    """Binarizes {0, 1} a real valued tensor."""

    def __init__(self, threshold=DEFAULT_THRESHOLD):
        super(Binarizer, self).__init__()
        self.threshold = threshold

    def forward(self, inputs):

arunmallya

class ElementWiseConv2d(nn.Module):
    """Modified conv. Do we need mask for biases too?"""

    def __init__(self, in_channels, out_channels, kernel_size, stride=1,
                 padding=0, dilation=1, groups=1, bias=True,
                 mask_init='1s', mask_scale=1e-2,
                 threshold_fn='binarizer', threshold=None):
        super(ElementWiseConv2d, self).__init__()
        kernel_size = _pair(kernel_size)
        stride = _pair(stride)

arunmallya

arunmallya