27 Matching Annotations
  1. Aug 2022
  2. May 2022
    1. For MacOS, you can do brew install git-lfs.

      Install for Mac: or pip install git-lfs or download from git-lfs website and run ./install.sh

    2. Let's also make a requirements.txt file which will allow us to install the packages that we need in whatever environment we need:

      requirements.txt crucial for libraries needed

  3. Mar 2017
    1. builtin = { 'settings' : ('settings', ), 'train' : ('train',),# 'training'), 'validate' : ('validate',),# 'validation'), 'test' : ('test', ), #'testing'), 'evaluate' : ('evaluate', ), #'evaluation'), 'templates' : ('templates', ), 'model' : ('model', ), 'loss' : ('loss', )

      to remove duplicated part

    1. num_entries = len(batch[keys[0]]) for entry in range(num_entries): print('Entry {}/{}:'.format(entry+1, num_entries)) for k in keys: print(' {}: {}'.format(k, batch[k][entry]))


    1. #### What is the shape of each key's values batch_shape = {} for k in keys: batch_shape[k] = batch[k].shape logger.info("A single batch consists of: %s", batch_shape) ### count inside the first element of batch dictionary, assign to num_entries num_entries = len(batch[keys[0]]) logger.info("There are total %s samples in a single batch.", num_entries) logger.info("Let's have a look a single sample") ### print out the whole batch, but one sample pair at a time, each sample pair like (key1:value, key2: value) # for entry in range(0, num_entries) ### Here only print out the first and last sample of a batch # for entry in range(0, num_entries, num_entries-1): for entry in range(1): ## print Entry entry_idx/total_entries print('Entry {}/{}:'.format(entry+1, num_entries)) ## for every batch_key for k in keys: # print: batch_key: this batch's first item print(' {}: {}'.format(k, batch[k][entry]))

      Hack kur data mnist.yml for simpler data overview

    1. Question: all models built successfully, but failed to train with train_network function. r2rt uses from tensorflow.models.rnn.ptb import reader which is not available anymore as I couldn't find it. I used reader.pyfrom https://github.com/tensorflow/models/tree/master/tutorials/rnn/ptb instead. I suspect the error lies on inappropriate use of functions in reader.py. Could anyone help me resolve it? Thanks! error message is here my guess on source of error Question: In order to upgrade CustomCell class to work, I need to upgrade the TF functions used by r2rt, e.g., tf.nn.rnn_cell._linear. However, I have difficulties to find the equivalent functions in TF docs. Could anyone help me update this class? Thanks codes that needs upgrade

      2 questions

    1. return (x, y)

      source of error may be here?

      • I tried use both return x, y and return (x,y)
      • both attempts made no differences
      • I guess it may be I used this reader.ptb_producer inappropriately by simply copying r2rt's code?
    1. # tf.nn.rnn_cell._linear can't be found??? ru0 = tf.nn.rnn_cell._linear([inputs, state]

      How to update this depreciated function, in order to make this CustomCell to work?

    2. TypeError

      Error message

  4. Feb 2017
    1. Batch size set to: 2 [INFO 2017-02-28 22:29:05,910 kur.providers.batch_provider:60] Maximum number of batches set to: 1

      I don't remember I set them as 2 and 1, what do them mean?

    2. Batch size set to: 32 [INFO 2017-02-28 22:29:02,413 kur.providers.batch_provider:60] Maximum number of batches set to: 50

      I set them to be 128 and 1000, why here are different?

    1. # Where to get the data cifar: &cifar url: "https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz" checksum: "6d958be074577803d12ecdefd02955f39262c83c16fe9348329d7fe0b5c001ce" path: "~/kur" # download and save dataset in kur directory, if so why not see it ######

      I can't find dataset inside ../kur/kur directory, where is cifar-10 stored?

      • home directory/kur
      • not working directory/kur
    2. images

      What does images refer to?

      • as cifar under settings gives no name as images
      • where does images come from?

      answer from adam

      Data suppliers are responsible for loading data and providing named data sources. So in the Kur tutorial, we use a Pickle data supplier to load the previously pickled data: a dictionary with keys "point" and "above", which are the names of the sources. These sources can then be referenced by name throughout the Kurfile. The CIFAR-10 data supplier is a "special" data supplier that "knows" how the CIFAR data is encoded, and loads the images and labels into named sources images and labels respectively. That's all--just a reference to the sources produced by the data supplier.

    1. by convolving with few 1x1 kernels, you are reducing the dimensions by collapsing many feature maps into few feature maps while maintaining the size of each feature map

      what is 1x1 conv really doing

    1. Hour 1: True transcript: these vast buildings what were they DNN prediction: he s ma tol ln wt r hett jzxzjxzjqzjqjxzq

      check how well the model after training one hour

  5. Jan 2017
    1. Copy either setup_p2.sh or setup_t2.sh to setup.sh (I did cp setup_t2.sh setup.sh on linux) Open setup.sh and change two lines -- change ami to the new ami that @vshets created, and change the VolumeSize to 30.

      How to reduce 128G EBS to 30G EBS

    1. The only working mobi2pdf website I have found so far. Also, Uploading and converting is fast and looks good

    1. distribution of mean 0 and variance 1 (if any of the are floats, they are first converted to integers by truncation)

      How to create a single random vector or matrix on mean 0 and variance 1

    1. Intuitively, an explanation is a local linear approximation of the model's behaviour

      LIME and Explanation

      Questions: Is LIME really resemble human's approach to understand black box of machine learning?

    1. [1],[[i*.0001,1-(i*.0001)]]

      This is the original code, but not working properly.

      please see the code and annotation above, where I made some changes and is plotting. However, I feel something not right about the plotting.

    2. [0,1 if i > 0 else 0],[[0.0001, 0.9999],[1-(i*.0001), i*.0001]]

      Here is what I altered in order to make the plotting work. But the plot looks not right, I don't know why. Could you have a look?


    1. Previous link won't take us to the highlighted text automatically because I shared the whole page link (top right) not the specific annotation link (bottom right).

      How to share specific annotation rather than a whole page

    2. annotation with jupyter notebook

      How to Annotate Jupyter notebook stored online