Ching-Chuan Chen's Blogger

Statistics, Machine Learning and Programming

0%

MxNet所提供的csv data iterator

Posted on Edited on In Disqus:

這篇是參考mxnet的一個example來的來源

而該篇所使用的資料是這一個比賽NDSB-II所提供的，請點我

登入Kaggle把Data裡面的四個zip下載下來

  1. sample_submission_validate.csv.zip
  2. train.csv.zip
  3. train.zip
  4. validate.zip

MxNet提供了一個Python去做Preprocessing

因為MxNet沒有提供R版本，所以我就寫了一版R的processing放在github上Repo連結

從Kaggle下載zip下來之後，放到一個資料夾(為了說明方便，這個資料夾就叫做root path)裡面

並從我的github repository clone下我建立的R專案，先跑preprocessing.R

接下來，我們就可以跑MxNet的train.R了：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
# Train.R for Second Annual Data Science Bowl
# Deep learning model with GPU support
# Please refer to https://mxnet.readthedocs.org/en/latest/build.html#r-package-installation
# for installation guide

require(mxnet)
require(data.table)

##A lenet style net, takes difference of each frame as input.
get.lenet <- function() {
  source <- mx.symbol.Variable("data")
  source <- (source-128) / 128
  frames <- mx.symbol.SliceChannel(source, num.outputs = 30)
  diffs <- list()
  for (i in 1:29) {
    diffs <- c(diffs, frames[[i + 1]] - frames[[i]])
  }
  diffs$num.args = 29
  source <- mxnet:::mx.varg.symbol.Concat(diffs)
  net <-
    mx.symbol.Convolution(source, kernel = c(5, 5), num.filter = 40)
  net <- mx.symbol.BatchNorm(net, fix.gamma = TRUE)
  net <- mx.symbol.Activation(net, act.type = "relu")
  net <-
    mx.symbol.Pooling(
      net, pool.type = "max", kernel = c(2, 2), stride = c(2, 2)
    )
  net <-
    mx.symbol.Convolution(net, kernel = c(3, 3), num.filter = 40)
  net <- mx.symbol.BatchNorm(net, fix.gamma = TRUE)
  net <- mx.symbol.Activation(net, act.type = "relu")
  net <-
    mx.symbol.Pooling(
      net, pool.type = "max", kernel = c(2, 2), stride = c(2, 2)
    )
  # first fullc
  flatten <- mx.symbol.Flatten(net)
  flatten <- mx.symbol.Dropout(flatten)
  fc1 <- mx.symbol.FullyConnected(data = flatten, num.hidden = 600)
  # Name the final layer as softmax so it auto matches the naming of data iterator
  # Otherwise we can also change the provide_data in the data iter
  return(mx.symbol.LogisticRegressionOutput(data = fc1, name = 'softmax'))
}

network <- get.lenet()
batch_size <- 32

# CSVIter is uesed here, since the data can't fit into memory
data_train <- mx.io.CSVIter(
  data.csv = "train-64x64-data.csv", data.shape = c(64, 64, 30),
  label.csv = "train-systole.csv", label.shape = 600,
  batch.size = batch_size
)

data_validate <- mx.io.CSVIter(
  data.csv = "validate-64x64-data.csv",
  data.shape = c(64, 64, 30),
  batch.size = 1
)

# Custom evaluation metric on CRPS.
mx.metric.CRPS <- mx.metric.custom("CRPS", function(label, pred) {
  pred <- as.array(pred)
  label <- as.array(label)
  for (i in 1:dim(pred)[2]) {
    for (j in 1:(dim(pred)[1] - 1)) {
      if (pred[j, i] > pred[j + 1, i]) {
        pred[j + 1, i] = pred[j, i]
      }
    }
  }
  return(sum((label - pred) ^ 2) / length(label))
})

# Training the stytole net
mx.set.seed(0)
stytole_model <- mx.model.FeedForward.create(
  X = data_train,
  ctx = mx.gpu(0),
  symbol = network,
  num.round = 65,
  learning.rate = 0.001,
  wd = 0.00001,
  momentum = 0.9,
  eval.metric = mx.metric.CRPS
)

# Predict stytole
stytole_prob = predict(stytole_model, data_validate)

# Training the diastole net
network = get.lenet()
batch_size = 32
data_train <-
  mx.io.CSVIter(
    data.csv = "./train-64x64-data.csv", data.shape = c(64, 64, 30),
    label.csv = "./train-diastole.csv", label.shape = 600,
    batch.size = batch_size
  )

diastole_model = mx.model.FeedForward.create(
  X = data_train,
  ctx = mx.gpu(0),
  symbol = network,
  num.round = 65,
  learning.rate = 0.001,
  wd = 0.00001,
  momentum = 0.9,
  eval.metric = mx.metric.CRPS
)

# Predict diastole
diastole_prob = predict(diastole_model, data_validate)

accumulate_result <- function(validate_lst, prob) {
  t <- read.table(validate_lst, sep = ",")
  p <- cbind(t[,1], t(prob))
  dt <- as.data.table(p)
  return(dt[, lapply(.SD, mean), by = V1])
}

stytole_result = as.data.frame(accumulate_result("./validate-label.csv", stytole_prob))
diastole_result = as.data.frame(accumulate_result("./validate-label.csv", diastole_prob))

train_csv <- read.table("./train-label.csv", sep = ',')

# we have 2 person missing due to frame selection, use udibr's hist result instead
doHist <- function(data) {
  res <- rep(0, 600)
  for (i in 1:length(data)) {
    for (j in round(data[i]):600) {
      res[j] = res[j] + 1
    }
  }
  return(res / length(data))
}

hSystole = doHist(train_csv[, 2])
hDiastole = doHist(train_csv[, 3])

res <- read.table("data/sample_submission_validate.csv", sep = ",", header = TRUE, stringsAsFactors = FALSE)

submission_helper <- function(pred) {
  for (i in 2:length(pred)) {
    if (pred[i] < pred[i - 1]) {
      pred[i] = pred[i - 1]
    }
  }
  return(pred)
}

for (i in 1:nrow(res)) {
  key <- unlist(strsplit(res$Id[i], "_"))[1]
  target <- unlist(strsplit(res$Id[i], "_"))[2]
  if (key %in% stytole_result$V1) {
    if (target == 'Diastole') {
      res[i, 2:601] <- submission_helper(diastole_result[which(diastole_result$V1 == key), 2:601])
    } else {
      res[i, 2:601] <- submission_helper(stytole_result[which(stytole_result$V1 == key), 2:601])
    }
  } else {
    if (target == 'Diastole') {
      res[i, 2:601] <- hDiastole
    } else {
      res[i, 2:601] <- hSystole
    }
  }
}

write.table(res, file = "submission.csv", sep = ",", quote = FALSE, row.names = FALSE)

本篇的重點在於下面這段R，用MxNet提供的mx.io.CSVIter去batch的訓練Net模型

而這裡的train-64x64-data.csv，每一行都是經過resized的30張圖片，所以data.shape是64 x 64 x 30

而label則每一行是長度600的binary vector，其shape設定成600

然後給好batch.size，MxNet就可以批次的從csv抓資料出來train模型了

不用一股腦地把資料全部匯入到R裡面再做，不然再多的記憶體也用不完Orz

1
2
3
4
5
data_train <- mx.io.CSVIter(
  data.csv = "train-64x64-data.csv", data.shape = c(64, 64, 30),
  label.csv = "train-systole.csv", label.shape = 600,
  batch.size = batch_size
)