Deep Networks with MNIST Data in R | Case Study with Keras & TensorFlow for Machine Intelligence

Поділитися
Вставка
  • Опубліковано 8 січ 2025

КОМЕНТАРІ • 72

  • @radosawpuakowski6982
    @radosawpuakowski6982 2 роки тому +1

    What and excellent piece of work! Introduction is complete, allowed me to understand and conduct whole analysis. Author comments build background knowledge and explain how to solve potential problems. @Dr. Bharatendra Rai my complements on your didactic skills and quality of materials!
    Btw I had to do minor script adaptions (probably due to some changes in library) to work in my environment.
    ln56: WAS: pred % predict_classes(testx) / IS: pred % predict(testx) %>% k_argmax()
    ln59: WAS: prob % predict_proba(testx) / IS: prob % predict(testx)
    Have fun!

    • @bkrai
      @bkrai  2 роки тому

      Thanks for the comments and code related update!

  • @farshidowrang2762
    @farshidowrang2762 3 роки тому +2

    The problem is that the codes "predict_classes" and "predict_proba" according to the Error massage are deprecated and they are removed from tensorflow in version 2.6 (I have the latest version). The Error Massage suggests "predict() %>% k_argmax()" and "predict()" instead but they are not working for me.

  • @lucasdemelo3
    @lucasdemelo3 5 років тому +1

    brilliant @
    Bharatendra Rai !

    • @bkrai
      @bkrai  5 років тому

      Thanks!

  • @scottsara123
    @scottsara123 5 років тому +2

    Sir waiting for video related to GAN Thanks for easy explain guidance to all self learner's

    • @bkrai
      @bkrai  5 років тому

      Thanks, I've it on my list!

  • @tarasst6887
    @tarasst6887 2 роки тому +1

    Great to see example with torch package

    • @bkrai
      @bkrai  2 роки тому

      Thanks!

  • @send2milan
    @send2milan 5 років тому

    Sir, you are the one who inspired me choosing analytics career. Thank for sharing all the videos in UA-cam.

    • @bkrai
      @bkrai  5 років тому

      Thanks for comments!

  • @adrianramos2989
    @adrianramos2989 5 років тому +1

    Terrific intro to the MNIST data set! Thank you for sharing such a quality content. I would like to ask you for the hand written pictures used in the example for replicability purposes. I've carefully followed your steps but have only achieved a very poor 30% accuracy when evaluating the MNIST trained model with my own handwritten numbers.

    • @bkrai
      @bkrai  5 років тому +1

      Here is the link:
      drive.google.com/open?id=1m_nKMCUZBP0O32Friuzck27QiiE9a7jz

  • @rebekahjardine8444
    @rebekahjardine8444 Рік тому +1

    If I have an entirely different set of images: A folder that contains a test and train folder with each containing two additional subfolders with labeled images. How can I replicate the same structure as the mnist dataset?

    • @bkrai
      @bkrai  Рік тому +1

      You can use the same approach that is used with my own handwritten images at the time of testing.

  • @flamboyantperson5936
    @flamboyantperson5936 5 років тому +1

    Great great tutorial on deep learning in R. Thank you very much Sir.

    • @bkrai
      @bkrai  5 років тому

      Thanks for comments!

  • @machinelearningzone.6230
    @machinelearningzone.6230 5 років тому +1

    Thanks for the tutorial and the code,but I faced the error"imensions must be equal, but are 32 and 320 for 'loss/dense_8_loss/softmax_cross_entropy_with_logits' (op: 'SoftmaxCrossEntropyWithLogits') with input shapes: [32,10], [320,10]." while fitting the model.
    Could you please help me out.

    • @bkrai
      @bkrai  4 роки тому

      Are your working with the same data/code used here? If yes, let me know which line you are referring to?

  • @gawine
    @gawine 9 місяців тому

    Hi! I followed every instruction given, but R studio still throws an error: No gradients provided for any variable. Does somebody have a clue why this occurs? Thanks!

  • @mannam5374
    @mannam5374 5 років тому +1

    Excellent video sir..your steps are very clear and easy to follow...please do a classification problem in deep learning keras using cloudml package....thanks Sir

    • @bkrai
      @bkrai  5 років тому

      Thanks for the suggestion, I've added this to my list.

  • @md.mahiuddin3097
    @md.mahiuddin3097 5 років тому +2

    This is a great lecture. Thank you so much.
    Sir, i want to predict a unlabeled data set by using labeled data set, then how can i proceed? Thanks in advance.

    • @bkrai
      @bkrai  5 років тому

      Train the network with labeled data and then use the model to predict from unlabeled data.

    • @md.mahiuddin3097
      @md.mahiuddin3097 5 років тому +1

      thank you @@bkrai

  • @JacksonUNT
    @JacksonUNT 2 роки тому

    Thank you for the video, but: 'Sequential' object has no attribute 'predict_classes'

  • @Rushield3981cc
    @Rushield3981cc 2 роки тому +1

    Dear Dr. Rai, I'm having an error while I try to upload my own images. "cannot allocate vector of size 2.9 Mb"
    Do you have any suggestion with this ?

    • @bkrai
      @bkrai  2 роки тому

      If you have any unnecessary data files loaded, you can remove them to create some space.

    • @Rushield3981cc
      @Rushield3981cc 2 роки тому +1

      @@bkrai Thank you for replying, I have no data before I load my 7000 image for classification . I have 16gb ram also.

    • @bkrai
      @bkrai  2 роки тому

      Try with less number of images or if possible, use gpu computer.

    • @Rushield3981cc
      @Rushield3981cc 2 роки тому

      @@bkrai Is it possible to use Gpu of our computer in R ? İnstead of cpu and ram ? (Thank you for replying)

  • @ambertrivedi1502
    @ambertrivedi1502 4 роки тому +1

    I'm getting an error that specified conda library does not exist

    • @bkrai
      @bkrai  4 роки тому

      I'll do a video to install keras and tensorflow in coming weeks.

    • @ambertrivedi1502
      @ambertrivedi1502 4 роки тому +1

      @@bkrai I already know how to install keras and tensorflow.
      1. #install packages
      install.packages("keras")
      2.load it into the library
      library("keras")
      3. You can look where it is loaded or not
      by typing
      search()
      #Search will give all loaded packages
      Run this command
      4.install_keras()
      #Another way is to install from github using devtools
      Since I'm running R in windows 10 I'm facing many issues.
      Python version 3.8.3
      Anaconda Latest Version
      Installed it with admin rights to configure it properly.
      Updated and upgraded pip and conda
      Installed Tensorflow CPU version with pip command from stackoverflow.
      Everthing is going well.
      But when I'm trying to load keras dataset I'm getting an error.
      Either python module so and so found or conda does not exist /Anaconda3/Scripts.
      This is where I'm struck.

    • @bkrai
      @bkrai  4 роки тому

      Probably it's something computer specific.

    • @ambertrivedi1502
      @ambertrivedi1502 4 роки тому

      @@bkrai Yes. I guess so. Sir you are indeed an amazing and inspiring teacher. Recently I had requested you to please do make a video on Plant disease detection. I hope you soon get it off your as well as my bucket list.

  • @ravikumar-rz8uu
    @ravikumar-rz8uu 5 років тому +1

    Very Helpful machine learning concept ,Thanks you ..
    can you make video for image processing in R.

    • @bkrai
      @bkrai  5 років тому

      Here is the playlist of related videos:
      ua-cam.com/play/PL34t5iLfZddsaZKDeCK2RD8lrhxBxp4hP.html

  • @ashishsharp
    @ashishsharp 4 роки тому +1

    Sir, Error in install_tensorflow() : could not find function "install_tensorflow", getting this error though I have tensor flow installed for Python. PN I am using R 3.6

    • @bkrai
      @bkrai  4 роки тому

      Try this:
      ua-cam.com/video/-IYYqdxdYXk/v-deo.html

  • @hammao
    @hammao 5 років тому +1

    Again thanks for referring me to this video... do you have any recording where you used flow_images_from_directory() to read in your images, the concept is still confusing to me because i couldn't find a way to look at the str of the object generated. Many Thanks .

    • @bkrai
      @bkrai  5 років тому

      I've not used flow_images_from_directory() so far.

    • @hammao
      @hammao 5 років тому +1

      Bharatendra Rai I figured it out already... many thanks

    • @bkrai
      @bkrai  5 років тому

      Thanks for the update!

  • @Felixinhk08
    @Felixinhk08 4 роки тому +1

    Hi, what is the activation ='relu' and 'softmax' means? Which one is better?

    • @kaihennig2323
      @kaihennig2323 3 роки тому +1

      Hi, I hope that this late answer still helps you.
      relu is a activation function, which puts everything below 0 to 0 and is linear from 0 to + infinity. Softmax uses the sigmoid (1 / (1+e^-x)) function but divides everything by the highest neuron in the output layer to have values between 0 and 1 which add up to 1.

    • @bkrai
      @bkrai  3 роки тому

      Thanks!

  • @johannaachhammer
    @johannaachhammer 4 роки тому +1

    Great introduction! Isn´t it a problem that accuracy for validation data is better than for training data when using dropout layers? Why don´t you reduce the epochs at the point where the loss function of the training data is at its lowest point? Thanks in advance.

    • @bkrai
      @bkrai  4 роки тому +1

      When difference is very small, it can happen sometimes.

    • @johannaachhammer
      @johannaachhammer 4 роки тому +1

      Thanks. So that is no Problem? It is unintuitive for me.
      And why don't you stop the Training ehen overfitting starts (after loss function of Validation Data reachs Minimum)?

    • @bkrai
      @bkrai  4 роки тому +1

      It's important to highlight the problem of over fitting and see how to identify it. And it is all the more important when it comes to deep learning.

  • @abhishekdas2640
    @abhishekdas2640 4 роки тому +1

    Sir, One error is showing while running the code
    "mnist

    • @bkrai
      @bkrai  4 роки тому +1

      Try this link:
      ua-cam.com/video/-IYYqdxdYXk/v-deo.html

    • @abhishekdas2640
      @abhishekdas2640 4 роки тому +1

      @@bkrai thank you verymuch sir. 😃

    • @bkrai
      @bkrai  4 роки тому

      Welcome!

  • @juanjosetorressanmartin1495
    @juanjosetorressanmartin1495 5 років тому +1

    can you help me?. every time when i want use mnist database. i have next error. please... i'm gonna go crazy.
    (Error in dir.exists(x) : file name conversion problem -- name too long?)
    this happens when I use keras library and I want to charge the data "mnist

  • @ravindarmadishetty736
    @ravindarmadishetty736 5 років тому +1

    Sir, expecting a video on recommendation system...thanks

    • @bkrai
      @bkrai  5 років тому

      Thanks, I've it on my list!

  • @ravikumar-rz8uu
    @ravikumar-rz8uu 5 років тому +1

    thank you, very nice response..

    • @bkrai
      @bkrai  5 років тому

      Thanks for comments!

  • @manjuyadavalwar3014
    @manjuyadavalwar3014 5 років тому +1

    Need your guidance as how we can solve sale prediction (Kaggle competition) . Below is problem snapshot - YYou are provided with daily historical sales data. The task is to forecast the total amount of products sold in every shop for the test set. Note that the list of shops and products slightly changes every month. Creating a robust model that can handle such situations is part of the challenge.
    File descriptions
    sales_train.csv - the training set. Daily historical data from January 2013 to October 2015.
    test.csv - the test set. You need to forecast the sales for these shops and products for November 2015.
    sample_submission.csv - a sample submission file in the correct format.
    items.csv - supplemental information about the items/products.
    item_categories.csv - supplemental information about the items categories.
    shops.csv- supplemental information about the shops.

    • @bkrai
      @bkrai  5 років тому

      Sorry I saw this today. I guess it's too late now.

  • @manjuyadavalwar3014
    @manjuyadavalwar3014 5 років тому +1

    Simplly Great :)

    • @bkrai
      @bkrai  5 років тому

      Thanks for comments!

  • @shahidtufail1009
    @shahidtufail1009 5 років тому +2

    not all heroes wear caps

    • @bkrai
      @bkrai  5 років тому

      thanks for comments!