I was searching for a solution on this great forum, but I didn't find it yet, so I am now writing what's my problem.
I am trying to dig into neural networks with R, by using multiple packages. The problem is the one from ...
[https://www.kaggle.com/c/titanic/overview]
With keras, I find a problem in the fitting step.
The model I built:
data$PassengerId <- NULL
data$Name <- NULL
data$Ticket <- NULL
data$Cabin <- NULL
library(caTools)
split = sample.split(data$Survived, SplitRatio = 0.75)
train_data <- subset(data, split==TRUE)
test_data <- subset(data, split==FALSE)
train_keras <- train_data
test_keras <- test_data
train_keras$Survived <- as.numeric(train_keras$Survived)
train_keras$Pclass <- as.numeric(train_keras$Pclass)
train_keras$Sex <- as.numeric(train_keras$Sex)
train_keras$Embarked <- as.numeric(train_keras$Embarked)
train_keras$Survived <- train_keras$Survived -1
test_keras$Survived <- as.numeric(test_keras$Survived)
test_keras$Pclass <- as.numeric(test_keras$Pclass)
test_keras$Sex <- as.numeric(test_keras$Sex)
test_keras$Embarked <- as.numeric(test_keras$Embarked)
test_keras$Survived <- test_keras$Survived -1
train_keras[,-1] <- scale(train_keras[,-1])
test_keras[,-1] <- scale(test_keras[,-1])
library(keras)
library(tidyverse)
X_train <- train_keras %>%
select(-Survived)
y_train <- to_categorical(train_keras$Survived)
X_test <- test_keras %>%
select(-Survived)
y_test <- to_categorical(test_keras$Survived)
classifier <- keras_model_sequential()
classifier %>%
layer_dense(units = 256, activation = 'relu', input_shape = ncol(X_train)) %>%
layer_dropout(rate = 0.4) %>%
layer_dense(units = 128, activation = 'relu') %>%
layer_dropout(rate = 0.3) %>%
layer_dense(units = 2, activation = 'sigmoid')
history <- classifier %>% compile(
loss = 'binary_crossentropy',
optimizer = 'adam',
metrics = c('accuracy')
)
So, after cleaning the data, dividing it into test and train, typing it correctly, I build the model in keras. Until now, no problem. The problem, as I said, it was in the moment of fitting this model.
classifier %>% fit(
X_train, y_train,
epochs = 100,
batch_size = 5,
validation_split = 0.2
)
This raises the error:
Error in py_call_impl(callable, dots$args, dots$keywords) :
ValueError: in user code:
Digging deeper, I see something strange, pointing at the dimensions on the sequential model.
assert_input_compatibility
' input tensors. Inputs received: ' +
str(inputs))
ValueError: Layer sequential_12 expects 1 inputs, but
it received 7 input tensors. Inputs received: [<tf.Tensor
'ExpandDims:0' shape=(None, 1) dtype=float32>
If I look even deeper, I see that the whole architecture was defined with tensors of (None,1) shape, which (and sorry for my lack of experience on this topic) for me is strange. I know the problem is the way I feed the model, but I still don't know when do I have the problems.
Thanks in advance
