stared / keras-sequential-ascii Goto Github PK

View Code? Open in Web Editor NEW

120.0 9.0 17.0 16 KB

ASCII summary for simple sequential models in Keras

Python 42.23% Jupyter Notebook 57.77%

ascii deep-learning convnet keras sequential-models keras-visualization

keras-sequential-ascii's Introduction

Sequential model in Keras -> ASCII

by Piotr Migdał

A library for Keras for investigating architectures and parameters of sequential models.

(discontinuted) For more general approaches, see: Simple diagrams of convoluted neural networks

Both model.summary() and graph export were not enough - I wanted array dimensions, numbers of parameters and activation functions in one place. I use it for didactic purpose.

TODO
- Add ASCII art for more layers.
- Go beyond simple sequential models (e.g. to allow merge layers); any ideas how?
- Consider PRing to the main Keras repo, see #3873.

See this library in the wild, for example:

Starting deep learning hands-on: image classification on CIFAR-10 - my post at deepsense.ai
Cifar-10 Classification using Keras Tutorial at Plon.io

Installation

From PyPI:

pip install keras_sequential_ascii

Or from this repo:

pip install git+git://github.com/stared/keras-sequential-ascii.git

Usage

from keras_sequential_ascii import keras2ascii
keras2ascii(model)

Examples

Proof of principle

           OPERATION           DATA DIMENSIONS   WEIGHTS(N)   WEIGHTS(%)

               Input   #####      3   32   32
  BatchNormalization    μ|σ  -------------------        64     0.1%
                       #####      3   32   32
       Convolution2D    \|/  -------------------       448     0.8%
                relu   #####     16   30   30
       Convolution2D    \|/  -------------------      2320     4.3%
                relu   #####     16   28   28
        MaxPooling2D   Y max -------------------         0     0.0%
                       #####     16   14   14
       Convolution2D    \|/  -------------------       272     0.5%
                tanh   #####     16   14   14
             Flatten   ||||| -------------------         0     0.0%
                       #####        3136
               Dense   XXXXX -------------------     50192    94.1%
                       #####          16
             Dropout    | || -------------------         0     0.0%
                       #####          16
               Dense   XXXXX -------------------        51     0.1%
             softmax   #####           3

VGG16

           OPERATION           DATA DIMENSIONS   WEIGHTS(N)   WEIGHTS(%)

              Input   #####      3  224  224
         InputLayer     |   -------------------         0     0.0%
                      #####      3  224  224
      Convolution2D    \|/  -------------------      1792     0.0%
               relu   #####     64  224  224
      Convolution2D    \|/  -------------------     36928     0.0%
               relu   #####     64  224  224
       MaxPooling2D   Y max -------------------         0     0.0%
                      #####     64  112  112
      Convolution2D    \|/  -------------------     73856     0.1%
               relu   #####    128  112  112
      Convolution2D    \|/  -------------------    147584     0.1%
               relu   #####    128  112  112
       MaxPooling2D   Y max -------------------         0     0.0%
                      #####    128   56   56
      Convolution2D    \|/  -------------------    295168     0.2%
               relu   #####    256   56   56
      Convolution2D    \|/  -------------------    590080     0.4%
               relu   #####    256   56   56
      Convolution2D    \|/  -------------------    590080     0.4%
               relu   #####    256   56   56
       MaxPooling2D   Y max -------------------         0     0.0%
                      #####    256   28   28
      Convolution2D    \|/  -------------------   1180160     0.9%
               relu   #####    512   28   28
      Convolution2D    \|/  -------------------   2359808     1.7%
               relu   #####    512   28   28
      Convolution2D    \|/  -------------------   2359808     1.7%
               relu   #####    512   28   28
       MaxPooling2D   Y max -------------------         0     0.0%
                      #####    512   14   14
      Convolution2D    \|/  -------------------   2359808     1.7%
               relu   #####    512   14   14
      Convolution2D    \|/  -------------------   2359808     1.7%
               relu   #####    512   14   14
      Convolution2D    \|/  -------------------   2359808     1.7%
               relu   #####    512   14   14
       MaxPooling2D   Y max -------------------         0     0.0%
                      #####    512    7    7
            Flatten   ||||| -------------------         0     0.0%
                      #####       25088
              Dense   XXXXX ------------------- 102764544    74.3%
               relu   #####        4096
              Dense   XXXXX -------------------  16781312    12.1%
               relu   #####        4096
              Dense   XXXXX -------------------   4097000     3.0%
            softmax   #####        1000

keras-sequential-ascii's People

Contributors

Stargazers

Watchers

Forkers

vyraun benjamesbabala libardo1 arkc hanfeijp simongeek stjordanis neveroldmilk b2220333 ashishpatel26 mehulfollytobevice lkampoli handsomewl fermiq ahmedyes2000 tyd19981111

keras-sequential-ascii's Issues

Error using lambda layers

           OPERATION           DATA DIMENSIONS   WEIGHTS(N)   WEIGHTS(%)

               Input   #####     72  300
          InputLayer     |   -------------------         0     0.0%
                       #####     72  300
               Dense   XXXXX -------------------     15050    92.0%
                       #####     72   50
          InputLayer     |   -------------------         0     0.0%
                       #####     72    3
         Concatenate   ????? -------------------         0     0.0%
                       #####     72   53
                LSTM   LLLLL -------------------      1180     7.0%
                tanh   #####           5
              Lambda   ????? -------------------         0     0.0%
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-222-96aac99b2755> in <module>()
----> 1 keras2ascii(lstm_model)

/anaconda3/envs/tensorflow_p27/lib/python2.7/site-packages/keras_sequential_ascii/main.pyc in sequential_model_to_ascii_printout(model, sparser, simplify, header)
    123 
    124 def sequential_model_to_ascii_printout(model, sparser=False, simplify=True, header=True):
--> 125     print_layers(jsonize(model), sparser=sparser, simplify=simplify, header=header)

/anaconda3/envs/tensorflow_p27/lib/python2.7/site-packages/keras_sequential_ascii/main.pyc in print_layers(jsonized_layers, sparser, simplify, header)
    118         print(data_template.format(
    119                 activation=each["activation"] if each["activation"] != "linear" else "",
--> 120                 shape=print_dim_tuple(each["output_shape"]),
    121                 # length=product(each["output_shape"])
    122         ))

/anaconda3/envs/tensorflow_p27/lib/python2.7/site-packages/keras_sequential_ascii/main.pyc in print_dim_tuple(t)
     83         return " ".join(["{:4d}".format(x) for x in t])
     84     else:
---> 85         return  "{:9d}".format(t[0])
     86 
     87 def print_layers(jsonized_layers, sparser=False, simplify=False, header=True):

IndexError: tuple index out of range

For model:

x = Input(shape=(72,300))
aux_input = Input(shape=(72, 3))
probs = Input(shape=(1,))
dim_reduct = Dense(50)(x)
cat = concatenate([dim_reduct, aux_input])
encoded = LSTM(5)(cat)
#cat2 = concatenate([encoded, probs])
#output = Dense(1, activation='linear')(encoded)

output = Lambda(lambda x: K.sum(x, axis=1))(encoded)

Batch Normalization layers not working?

Even though your example in the readme.md runs BN layers, it seems that I'm having issues with them anyway.

Info: I use the latest tensorflow gpu version (1.1.0) and Keras v2.0.4, under python 2.7.

Error code:

  File "/usr/local/lib/python2.7/dist-packages/keras_sequential_ascii/main.py", line 123, in sequential_model_to_ascii_printout
    print_layers(jsonize(model), sparser=sparser, simplify=simplify, header=header)
  File "/usr/local/lib/python2.7/dist-packages/keras_sequential_ascii/main.py", line 110, in print_layers

    percent_parameters=100 * each["n_parameters"] / all_weights
UnicodeEncodeError: 'ascii' codec can't encode characters in position 24-25: ordinal not in range(128)

My guess is that the characters μ and σ displayed (that ran fine for you somehow) can't get encoded for me... How could we solve this?

EDIT: I sort of fixed it by digging into main.py (edited with nano): I found out that the template associated with BN layers were gibberish (mainly blank, with a single ^ at the end) whereas it should've been μ|σ. I just replaced it with m and v (mean and var) I guess it'll work for now

Export to PNG diagram too?

Congrats @stared for this cool project!
Do you think you could do a version that outputs such 3D diagrams too?

Or do you know another library that takes a Keras model object as input, and that can output such diagrams?

stared / keras-sequential-ascii Goto Github PK

keras-sequential-ascii's Introduction

Sequential model in Keras -> ASCII

Installation

Usage

Examples

Proof of principle

VGG16

keras-sequential-ascii's People

Contributors

Stargazers

Watchers

Forkers

keras-sequential-ascii's Issues

Error using lambda layers

Batch Normalization layers not working?

Export to PNG diagram too?

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent