如何使 input_shape 参数适应 R 中的表格尺寸

Question

到目前为止，我的网络是这样的：

data <- data2018

# Splitting data into train and test
set.seed(101)
sample <- sample.int(n = nrow(data), size = floor(.75 * nrow(data)), replace = F)
train <- as.data.table(data[sample, ])
test <- as.data.table(data[-sample, ])

# Preparing the training set
cols <- c(names(train))
train %<>% mutate_at(cols, as.factor)
train %<>% mutate_at(cols, as.numeric)
train %<>% mutate_at(cols,~(scale(.) %>% as.vector))
train[is.na(train) ] <- 0

# Splitting train into x (parameters) and y (prediction values)
x_train <- data.matrix(train[,-"1"])
y_train <- data.matrix(train[, "1"])

# Designing model
model <- keras_model_sequential() %>%
layer_dense(units = 128, activation = 'tanh', input_shape = dim(x_train)[-1]) %>%
layer_dropout(rate = 0.4) %>%
layer_dense(units = 64, activation = 'tanh') %>%
layer_dropout(rate = 0.3) %>%
layer_dense(units = 32, activation = 'tanh') %>%
layer_dropout(rate = 0.2) %>%
layer_dense(units = 16, activation = 'tanh') %>%
layer_dropout(rate = 0.1) %>%
layer_dense(units = length(unique(data$`1`)), activation = 'softmax')

# Summarize model layers and units
summary(model)

# Compile model
model %>% compile(
loss = 'categorical_crossentropy',
optimizer = optimizer_rmsprop(),
metrics = c('accuracy')
)

# Train model
history <- model %>% fit(
x_train, y_train,
epochs = 100, batch_size = 32,
validation_split = 0.2
)

由于保密问题，我无法显示原始数据表，但它是一个由 7 行和 34 列组成的简单表（只是一个小的模拟块）。

输入 (x_train) 已经标准化和矢量化。这是一个包含行和列的简单表格。

dim(x_train)

结果：

> dim(x_train)
[1]  5 33

但是，如果我尝试训练模型，keras 会打印出这条神秘的错误消息。

Error in py_call_impl(callable, dots$args, dots$keywords) : 
ValueError: in user code:

/home/seschlbeck/.local/share/r-miniconda/envs/r-reticulate/lib/python3.6/site-   packages/tensorflow/python/keras/engine/training.py:855 train_function  *
    return step_function(self, iterator)
/home/seschlbeck/.local/share/r-miniconda/envs/r-reticulate/lib/python3.6/site-packages/tensorflow/python/keras/engine/training.py:845 step_function  **
    outputs = model.distribute_strategy.run(run_step, args=(data,))
/home/seschlbeck/.local/share/r-miniconda/envs/r-reticulate/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:1285 run
    return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
/home/seschlbeck/.local/share/r-miniconda/envs/r-reticulate/lib/python3.6/site-packages/tensorflow/python/distribute/distribute_lib.py:2833 call_for_each_replica
    return self._call_for_each_replica(fn, args, kwargs)
/home/seschlbeck/.local/share/r-miniconda/envs/r-reticulate/lib/python3

我猜这与我的输入形状有关。但我就是想不通为什么... 我已经尝试了以下几种（都不起作用）

dim(x_train)
c(5, 33)
c(None, 5, 33)
dim(x_train)[1]

有人知道我在这里做错了什么吗？任何帮助表示赞赏...

Answer 1

input_shape 应该没问题。当“输出”层中的单元数指定错误时，会出现这种错误消息。您的训练样本中可能只有两个类别，即二进制分类，但您的代码适用于 n-class 案例，您需要 n 单位和分类交叉熵损失。尝试在最后一个密集层中设置一个单个单元：

layer_dense(units = 1, activation = 'softmax')