-
Notifications
You must be signed in to change notification settings - Fork 1
/
kaggle_keras.R
160 lines (128 loc) · 4.67 KB
/
kaggle_keras.R
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
## libraries
library(tidyverse)
library(caret)
library(here)
library(EBImage)
library(doParallel)
library(reticulate)
library(keras)
library(tensorflow)
library(tidymodels)
## data
train_path <- './train'
test_path <- './test'
#train_path <- here("data","train")
#test_path <- here("data", "test")
## Cluster
nCores<- detectCores()-1
registerDoParallel(makeCluster(nCores))
## Image processing
proc_img<- function(file){
image<- readImage(file)
#image.cl<- clahe(image)
#image.n<- normalize(image.cl)
array_reshape(image,c(32,32,3))
}
edge3<- function(df, dims){
res1<- NULL
cvar<- dims[2]
res1<- foreach(i = 1:nrow(df), .combine = rbind) %dopar% {
ar<- array(df[i,], dim = dims)
edgs<- array(cbind((ar[,1:(cvar-1)] - ar[,2:cvar]), rep(0, dims[1])))
# if(i%%100 == 0){print(paste("Observation: ",i))}
}
res2<- NULL
cvar<- dims[2]
res2<- foreach(i = 1:nrow(df), .combine = rbind) %dopar% {
ar<- array(df[i,], dim = dims)
edgs<- array(rbind((ar[1:(cvar-1),] - ar[2:cvar,]), rep(0, dims[2])))
# if(i%%100 == 0){print(paste("Observation: ",i))}
}
res<- cbind(res1,res2)
return(res)
}
## load training data
train_csv <- read.csv("../input/aerial-cactus-identification/train.csv")
test_labels<- data.frame(id = list.files(test_path))
#train_csv <- read.csv(here("data", "train.csv"))
#test_labels<- data.frame(id = list.files(here("data", "test")))
## training split
training<- createDataPartition(train_csv$has_cactus, p=0.8, list = F)
train_imgs<- paste(train_path, train_csv$id[training], sep = '/')
valid_imgs<- paste(train_path, train_csv$id[-training], sep = '/')
test_imgs<- paste(test_path, test_labels$id, sep = '/')
## answers split
train_y<- train_csv$has_cactus[training]
valid_y<- train_csv$has_cactus[-training]
## Load training images
dims <- length(train_imgs)
train_x = NULL
train_x<- foreach(i = 1:dims, .packages = c("EBImage","reticulate"), .combine = rbind) %dopar% {
train_x[[i]] <- proc_img(train_imgs[i])
}
dims <- length(valid_imgs)
valid_x = NULL
valid_x<- foreach(i = 1:dims, .packages = c("EBImage","reticulate"), .combine = rbind) %dopar% {
valid_x[[i]] <- proc_img(valid_imgs[i])
}
## test images
dims <- length(test_imgs)
test_x = NULL
test_x<- foreach(i = 1:dims, .packages = c("EBImage","reticulate"), .combine = rbind) %dopar% {
test_x[[i]] <- proc_img(test_imgs[i])
}
## Edge features
dims<- c(32,32)
ed1.t<- edge3(train_x[,1:1024], dims)
ed2.t<- edge3(train_x[,1025:2048], dims)
ed3.t<- edge3(train_x[,2049:3072], dims)
ed1.v<- edge3(valid_x[,1:1024], dims)
ed2.v<- edge3(valid_x[,1025:2048], dims)
ed3.v<- edge3(valid_x[,2049:3072], dims)
ed1.test<- edge3(test_x[,1:1024], dims)
ed2.test<- edge3(test_x[,1025:2048], dims)
ed3.test<- edge3(test_x[,2049:3072], dims)
train_x<- cbind(train_x, ed1.t, ed2.t, ed3.t)
valid_x3<- cbind(valid_x, ed1.v, ed2.v, ed3.v)
test_x<- cbind(test_x, ed1.test, ed2.test, ed3.test)
## keras CNN
train_x_reshape = array_reshape(x = train_x, dim = list(14000, 32, 32, 9))
valid_x_reshape = array_reshape(x = valid_x, dim = list(3500, 32, 32, 9))
test_x_reshape = array_reshape(x = test_x, dim = list(4000, 32, 32, 9))
## Response variable (hot coding)
trainLabels= to_categorical(train_y)
validLabels= to_categorical(valid_y)
# Create Model
model <- keras_model_sequential()
model %>%
layer_conv_2d(filters = 32, kernel_size = c(3,3), activation = "relu",
input_shape = c(32,32,9)) %>%
layer_max_pooling_2d(pool_size = c(2,2)) %>%
layer_conv_2d(filters = 64, kernel_size = c(3,3), activation = "relu") %>%
layer_max_pooling_2d(pool_size = c(2,2)) %>%
layer_conv_2d(filters = 64, kernel_size = c(3,3), activation = "relu")
model %>%
layer_flatten() %>%
layer_dense(units = 64, activation = "relu") %>%
layer_dense(units = 2, activation = "softmax")
#summary(model)
## Compile Model
model %>%
compile(loss = 'binary_crossentropy',
optimizer = optimizer_adam(),
#optimizer = optimizer_rmsprop(lr = 0.0001, decay = 1e-6),
metrics = 'accuracy')
## Fit the model
fitModel <- model %>%
fit(train_x_reshape,
trainLabels,
epochs = 50, # one pass over the entire dataset, evalutate at end of each epoch
batch_size = 500, # number of samples processed independently, in parallel. 1 update/batch
validation_data = list(valid_x_reshape,validLabels)
)
## Results (table)
pred.prob <- model %>% predict(test_x_reshape)
preds<- as.numeric(pred.prob[,2]>0.7 )
submission<- data.frame(id=test_labels$id , has_cactus=preds)
str(submission)
write_csv(submission, "submission.csv")