我有一个时间序列数据集,其中包含3个测量变量和大约2000个样本。我想使用RNN或Keras在R中使用RNN或1D CNN模型将样本分为4个类别中的1个。我的问题是我无法成功重塑k_reshape()
函数的模型。
我沿着Ch。 Chollet&Allaire的 R的深度学习中的第6章,但是他们的示例与我的数据集相差无几,因此我现在感到困惑。我试图模仿本书那章中的代码,但毫无用处。 Here's a link to the source code for the chapter.
library(keras)
df <- data.frame()
for (i in c(1:20)) {
time <- c(1:100)
var1 <- runif(100)
var2 <- runif(100)
var3 <- runif(100)
run <- data.frame(time, var1, var2, var3)
run$sample <- i
run$class <- sample(c(1:4), 1)
df <- rbind(df, run)
}
head(df)
# time feature1 feature2 feature3 sample class
# 1 0.4168828 0.1152874 0.0004415961 1 4
# 2 0.7872770 0.2869975 0.8809415097 1 4
# 3 0.7361959 0.5528836 0.7201276931 1 4
# 4 0.6991283 0.1019354 0.8873193581 1 4
# 5 0.8900918 0.6512922 0.3656302236 1 4
# 6 0.6262068 0.1773450 0.3722923032 1 4
k_reshape(df, shape(10, 100, 3))
# Error in py_call_impl(callable, dots$args, dots$keywords) :
# TypeError: Failed to convert object of type <class 'dict'> to Tensor. Contents: {'time': [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, 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, 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, 3
我对重塑数组非常陌生,但是我想拥有一个形状为(samples, time, features)
的数组。我很想听听有关如何正确重塑该数组的建议,或者有关如果我在这方面的基础不足的话,有关如何为DL模型处理该数据的指南。
答案 0 :(得分:0)
我找到了两个解决方案。我的困惑来自k_reshape
的错误消息,我不理解该如何解释。
array_reshape()
函数。k_reshape()
函数,但这一次使用适当的形状。这是我成功执行的代码:
# generate data frame
dat <- data.frame()
for (i in c(1:20)) {
time <- c(1:100)
var1 <- runif(100)
var2 <- runif(100)
var3 <- runif(100)
run <- data.frame(time, var1, var2, var3)
run$sample <- i
run$class <- sample(c(1:4), 1)
dat <- rbind(df, run)
}
dat_m <- as.matrix(df) # convert data frame to matrix
# time feature1 feature2 feature3 sample class
# 1 0.4168828 0.1152874 0.0004415961 1 4
# 2 0.7872770 0.2869975 0.8809415097 1 4
# 3 0.7361959 0.5528836 0.7201276931 1 4
# 4 0.6991283 0.1019354 0.8873193581 1 4
# 5 0.8900918 0.6512922 0.3656302236 1 4
# 6 0.6262068 0.1773450 0.3722923032 1 4
# solution with reticulate's array_reshape function
dat_array <- reticulate::array_reshape(x = dat_m[,c(2:4)], dim = c(20, 100, 3))
dim(dat_array)
# [1] 20 100 3
class(dat_array)
# [1] "array"
# solution with keras's k_reshape
dat_array_2 <- keras::k_reshape(x = dat_m[,c(2:4)], shape = c(20, 100, 3))
dim(dat_array)
# [1] 20 100 3
class(dat_array)
# [1] 20 100 3
class(dat_array_2)
# [1] "tensorflow.tensor" "tensorflow.python.framework.ops.Tensor"
# [3] "tensorflow.python.framework.ops._TensorLike" "python.builtin.object"
一些注意事项:
array_reshape
的输出是一个数组类,但是k_reshape()
输出一个tensorflow张量对象。两者都在创建的深度学习网络中为我工作,但是我发现数组类更具解释性。