Deep learning in R | Arpan Gupta | Data Scientist & IITian



● Learn R/Python programming /data science /machine learning/AI
● Wants to know R /Python code
● Wants to learn about decision tree,random forest,deeplearning,linear regression,logistic regression,H2o,neural network,Xgboost, gbm, bagging and so in R/Python?
● Wants to become a data scientist using R/Python ?
● Wants to be a leader in business or data analytics using R/Python ?
● Know about latest analytics techniques including graph theory,deeplearning, social network analytics,…

, , , , , , , ,

13 thoughts on “Deep learning in R | Arpan Gupta | Data Scientist & IITian

  1. can't i use relu for hidden layers and softmax in output layer ?

    act.fct only have tanh and sigmoid function in it … i tried using —

    softmax <- function (x) {exp(x) – logsumexp(x)} [Error: ''act.fct' is not known ]

    is there any ways where i can use different activation function in different layers rather than using just sigmoid or tanh in every layer of my model???

  2. 1) How do I treat categorical independent variables while performing a regression using deep learning?
    2) How do I come to know the significance(p-value) of the predictor variables in my model?

  3. Hi Arpan thanks. Trying to run the same code of yours but getting the error as Error:
    unexpected argument "distribution", is this legacy code? Try ?h2o.shim..please have a look and help.. Here is my code:;h2o.init()
    iris=as.h2o(iris)
    iris$Species <- as.factor(iris$Species) #Encode response as categorical
    split <- h2o.splitFrame(data = iris, ratios = 0.7)
    train <- split[[1]]
    test <- split[[2]]
    dim(train)
    dim(test)
    y <- "Species"
    x <- setdiff(names(train), y)

    h2o.shim(enable = TRUE)
    #deep learning model
    train=na.omit(train)
    fit2 <- h2o.deeplearning(x = x, y = y, seed=1234,
    training_frame = as.h2o(train),activation="Tanh",hidden=c(10,10),loss="CrossEntropy",distribution="AUTO")

  4. I am using this in a university assignment and they have stated "Build an autoencoder with only one hidden layer and change the number of its neurons from 2 to 50.
    What parameters in h2o.deeplearning do i change?

Leave a Reply

Your email address will not be published. Required fields are marked *