├── README.md
├── arch.png
├── net.cfg
└── nnplot.py


/README.md:
--------------------------------------------------------------------------------
1 | # nnplot
2 | python script for plotting various neural network architectures
3 | 
4 | for now, edit the "io" and "layers" variables in nnplot.py
5 | 
6 | Some example output ...
7 | ![Alt text](arch.png?raw=true "Title")
8 | 


--------------------------------------------------------------------------------
/arch.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/osh/nnplot/cadf6a182f169d4c9bd403c300fd7282fa513972/arch.png


--------------------------------------------------------------------------------
/net.cfg:
--------------------------------------------------------------------------------
 1 | {
 2 |     "io": [
 3 |         [ "Input", [ 2, 128 ] ],
 4 |         [ "Output", [ 2, 128 ] ]
 5 |     ],
 6 |     "layers": [
 7 |         [ "Conv2D(1,1,40)\nLinear", 0.5, 40, 5 ],
 8 |         [ "Dense(44)\nRelu", null, 10, 2 ],
 9 |         [ "Dense(2*88)\nRelu", null, 15, 3 ],
10 |         [ "Conv2D(1,1,81)\nLinear", 0.5, 40, 5 ]
11 |     ]
12 | }
13 | 


--------------------------------------------------------------------------------
/nnplot.py:
--------------------------------------------------------------------------------
 1 | #!/usr/bin/env python
 2 | #
 3 | # NN Plot
 4 | # Tim O'Shea (c) 2016
 5 | #
 6 | #
 7 | #
 8 | import matplotlib.pyplot as plt
 9 | import matplotlib.patches as patches
10 | import matplotlib.lines as lines
11 | 
12 | # load the config def
13 | import sys,json
14 | fn = sys.argv[1]
15 | print fn
16 | cfg = json.loads(open(fn).read())
17 | io = cfg["io"]
18 | layers = cfg["layers"]
19 | 
20 | # ...
21 | maxy = max(map(lambda x: x[3], layers))
22 | print maxy
23 | fig = plt.figure()
24 | ax = fig.add_subplot(111)
25 | for i,l in enumerate(layers):
26 |     ll=0.25 + i
27 |     w = 0.5
28 |     h = layers[i][3]
29 |     lr=(maxy-h)/2.0
30 | 
31 |     print (ll,lr,w,h)
32 |     ax.add_patch(
33 |         patches.Rectangle( (ll, lr), w, h, fill=False)
34 |         )
35 |     ax.annotate(l[0], xy=(ll+w/2.0, lr+h/2.0),
36 |         ha='center', va='center',
37 |         rotation=90)
38 | 
39 |     if(i < len(layers)-1):
40 | 
41 |         h_n = layers[i+1][3]
42 |         lr_n=(maxy-h_n)/2.0
43 |         i_x = ll+w
44 |         o_x = ll+1
45 | 
46 |         for j in range(0,layers[i][2]):
47 |             for k in range(0,layers[i+1][2]):
48 |                 j_rel = (j*1.0/(layers[i][2]-1))
49 |                 k_rel = (k*1.0/(layers[i+1][2]-1))
50 |                 if(l[1]==None or abs(j_rel - k_rel)<l[1]):
51 |                     i_ht = lr+h*j_rel
52 |                     o_ht = lr_n+h_n*k_rel
53 |                     ax.add_line(lines.Line2D([i_x,o_x], [i_ht,o_ht], color='black', linestyle='solid'))
54 | 
55 |         if(i == 0):
56 |             for j in range(0,l[2]):
57 |                 j_rel = (j*1.0/(l[2]-1))
58 |                 ax.add_patch( patches.Circle([i_x-w,lr+h*j_rel], 0.025, color='blue', ec="none") )
59 | 
60 |         if(i == len(layers)-2):
61 |             for j in range(0,layers[i+1][2]):
62 |                 j_rel = (j*1.0/(layers[i+1][2]-1))
63 |                 ax.add_patch( patches.Circle([o_x+w,lr_n+h_n*j_rel], 0.025, color='blue', ec="none") )
64 | 
65 | ax.annotate("Inputs (2x88)", xy=(0,maxy/2.0),
66 |         ha='center',va='center',
67 |         rotation=90)
68 | ax.annotate("Outputs (2x88)", xy=(len(layers),maxy/2.0),
69 |         ha='center',va='center',
70 |         rotation=90)
71 | 
72 | ymarg = 0.05
73 | xmarg = 0.05
74 | plt.ylim(-ymarg, maxy+0.5+ymarg)
75 | plt.xlim(-xmarg,len(layers)+xmarg)
76 | plt.axis('off')
77 | plt.savefig('test.png', bbix_inches='tight')
78 | plt.show()
79 | 


--------------------------------------------------------------------------------