ValueError: in user code while using keras model in R

Question

I am trying to run 1 dimensional CNN in R using keras package. I am using the following code

library(MASS)
library(keras)

##Create some data
data("Boston")
data <- Boston
# create a list of 70% of the rows in the original dataset we can use for training
set.seed(123)
training <- sample(nrow(data), 0.7 * nrow(data))

dataTrain <- data[training,]
dataTest <- data[-training,]

dataTrain_y <- as.matrix(dataTrain$medv)
dataTrain_x <- as.matrix(subset(dataTrain, select = -c(medv)))

dataTest_y <- as.matrix(dataTest$medv)
dataTest_x <- as.matrix(subset(dataTest, select = -c(medv)))

#Reshaping the data for CNN
dataTrain_x <- array_reshape(dataTrain_x, c(ncol(dataTrain_x), nrow(dataTrain_x), 1))
dataTest_x <- array_reshape(dataTest_x, c(ncol(dataTest_x), nrow(dataTest_x), 1))

#CNN model
model <- keras_model_sequential() %>%
  layer_conv_1d(filters=32, kernel_size=4, activation="relu", 
                input_shape=c(ncol(dataTrain_x), nrow(dataTrain_x))) %>%
  layer_max_pooling_1d(pool_size=2) %>%
  layer_conv_1d(filters=64, kernel_size=2, activation="relu") %>%
  layer_max_pooling_1d(pool_size=2) %>%
  layer_dropout(rate=0.4) %>%
  layer_flatten() %>%
  layer_dense(units=100, activation="relu") %>%
  layer_dropout(rate=0.2) %>%
  layer_dense(units=1, activation="linear")

model %>% compile(
  loss = "mse",
  optimizer =  "adam", #'sgd' can also be used
  metrics = list("mean_absolute_error")
)

model %>% summary()

history <- model %>% fit(dataTrain_x, dataTrain_y, 
                         epochs = 100, batch_size = 50, 
                         #callbacks = callback_tensorboard("logs/run_a"),
                         validation_split = 0.2)

But it returns me the following error

Error in py_call_impl(callable, dots$args, dots$keywords) : 
  ValueError: in user code:
    C:\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py:805 train_function  *
        return step_function(self, iterator)
    C:\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py:795 step_function  **
        outputs = model.distribute_strategy.run(run_step, args=(data,))
    C:\Python37\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:1259 run
        return self._extended.call_for_each_replica(fn, args=args, kwargs=kwargs)
    C:\Python37\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:2730 call_for_each_replica
        return self._call_for_each_replica(fn, args, kwargs)
    C:\Python37\lib\site-packages\tensorflow\python\distribute\distribute_lib.py:3417 _call_for_each_replica
        return fn(*args, **kwargs)
    C:\Python37\lib\site-packages\tensorflow\python\keras\engine\training.py:788 run_step  **
        outputs = model.train_step(data)
    C:\Python37\lib\site-pac

Now how can I solve this error?

jared_mamrot · Accepted Answer · 2021-07-20 03:20:56Z

2

+50

The error message is truncated, so it's not much help, but it looks to me like you've got some minor typos. If you make the suggested changes (the "###" comments) the model appears to compile and train as expected, e.g.

library(MASS)
library(keras)

##Create some data
data("Boston")
data <- Boston
# create a list of 70% of the rows in the original dataset we can use for training
set.seed(123)
training <- sample(nrow(data), 0.7 * nrow(data))

dataTrain <- data[training,]
dataTest <- data[-training,]

dataTrain_y <- as.matrix(dataTrain$medv)
dataTrain_x <- as.matrix(subset(dataTrain, select = -c(medv)))

dataTest_y <- as.matrix(dataTest$medv)
dataTest_x <- as.matrix(subset(dataTest, select = -c(medv)))

#Reshaping the data for CNN
### These dimensions don't look correct; switch ncol() with nrow()
dataTrain_x <- array_reshape(dataTrain_x, c(nrow(dataTrain_x), ncol(dataTrain_x), 1))
dataTest_x <- array_reshape(dataTest_x, c(nrow(dataTest_x), ncol(dataTest_x), 1))

#CNN model
model <- keras_model_sequential() %>%
  layer_conv_1d(filters=32, kernel_size=4, activation="relu", 
                ### The input shape doesn't look correct; instead of 
                ### `c(ncol(dataTrain_x), nrow(dataTrain_x))` (354, 13)
                ### I believe you want `dim(dataTest_x)` (13, 1)
                input_shape=c(ncol(dataTrain_x), 1)) %>%
  layer_max_pooling_1d(pool_size=2) %>%
  layer_conv_1d(filters=64, kernel_size=2, activation="relu") %>%
  layer_max_pooling_1d(pool_size=2) %>%
  layer_dropout(rate=0.4) %>%
  layer_flatten() %>%
  layer_dense(units=100, activation="relu") %>%
  layer_dropout(rate=0.2) %>%
  layer_dense(units=1, activation="linear")

model %>% compile(
  loss = "mse",
  optimizer =  "adam", #'sgd' can also be used
  metrics = list("mean_absolute_error")
)

model %>% summary()
#> Model: "sequential"
#> ________________________________________________________________________________
#> Layer (type)                        Output Shape                    Param #     
#> ================================================================================
#> conv1d_1 (Conv1D)                   (None, 10, 32)                  160         
#> ________________________________________________________________________________
#> max_pooling1d_1 (MaxPooling1D)      (None, 5, 32)                   0           
#> ________________________________________________________________________________
#> conv1d (Conv1D)                     (None, 4, 64)                   4160        
#> ________________________________________________________________________________
#> max_pooling1d (MaxPooling1D)        (None, 2, 64)                   0           
#> ________________________________________________________________________________
#> dropout_1 (Dropout)                 (None, 2, 64)                   0           
#> ________________________________________________________________________________
#> flatten (Flatten)                   (None, 128)                     0           
#> ________________________________________________________________________________
#> dense_1 (Dense)                     (None, 100)                     12900       
#> ________________________________________________________________________________
#> dropout (Dropout)                   (None, 100)                     0           
#> ________________________________________________________________________________
#> dense (Dense)                       (None, 1)                       101         
#> ================================================================================
#> Total params: 17,321
#> Trainable params: 17,321
#> Non-trainable params: 0
#> ________________________________________________________________________________

history <- model %>% fit(dataTrain_x, dataTrain_y, 
                         epochs = 100, batch_size = 50, 
                         #callbacks = callback_tensorboard("logs/run_a"),
                         validation_split = 0.2)

^{Created on 2021-07-20 by the reprex package (v2.0.0)}

edited Jul 20, 2021 at 3:20

answered Jul 20, 2021 at 0:17

jared_mamrot

26.5k5 gold badges27 silver badges56 bronze badges

Sign up to request clarification or add additional context in comments.

6 Comments

UseR10085 Over a year ago

Thank you now the model runs as you have said. But when I try to run further code like model %>% evaluate(dataTrain_x, dataTrain_y, verbose = 0) or model %>% predict(dataTest_x), it gives error like

Error in py_call_impl(callable, dots$args, dots$keywords) :    ValueError: Data cardinality is ambiguous:   x sizes: 13   y sizes: 354 Make sure all arrays contain the same number of samples.

Can you please check it?

jared_mamrot Over a year ago

Both of those commands work as expected on my system: evaluate() returns "loss 103.927383" and "mean_absolute_error 7.386703" and predict() returns a column of probabilities. Is this a problem with your 'actual' data, and/or with the Boston example data? Perhaps remove all objects from the environment and try again?

UseR10085 Over a year ago

I have copied your full code and ran it. It is now running. But I am getting different loss and mean_absolute_error. Why this may be happening? Can you explain how to decide the input_shape?

jared_mamrot Over a year ago

You get different values for loss/MAE because your model is different to mine. This behaviour is expected, because CNN training is stochastic (machinelearningmastery.com/stochastic-in-machine-learning). The input_shape is explained: rdocumentation.org/packages/keras/versions/2.4.0/topics/…. Basically, it's the "Dimensionality of the input not including the samples axis" , so for your array (354, 13, 1) you exclude the samples axis (354) and you have (13, 1) (13 vectors of 1-dimensional vectors). See baeldung.com/cs/ml-understanding-dimensions-cnn

UseR10085 Over a year ago

Thank you very much for the informative comment. There is no way like using set.seed() to get the same output for the same input set?

|

Collectives™ on Stack Overflow

ValueError: in user code while using keras model in R

1 Answer 1

6 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

6 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related