python
/
nn-from-scratch
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Update nn-from-scratch.ipynb 

Found a typo in the backpropagation formula in the ipython notebook file on line 290. It should be 1 - tanh^2(z1), rather than 1-tanh^2(z2). It is correctly implemented in the code below and even in the description in the blog post: http://www.wildml.com/2015/09/implementing-a-neural-network-from-scratch/. 

Also added subscripts to delta_2 on lines 293 and 294.
pull/3/head
jgdwyer 2016-04-27 14:29:13 -04:00
parent 9f4a767542
commit 5b69dbd5ce
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
nn-from-scratch.ipynb
View File

@ -287,11 +287,11 @@
"$$\n",
"\\begin{aligned}\n",
"& \\delta_3 = y - \\hat{y} \\\\\n",
"& \\delta_2 = (1 - \\tanh^2z_2) \\circ \\delta_3W_2^T \\\\\n",
"& \\delta_2 = (1 - \\tanh^2z_1) \\circ \\delta_3W_2^T \\\\\n",
"& \\frac{\\partial{L}}{\\partial{W_2}} = a_1^T \\delta_3 \\\\\n",
"& \\frac{\\partial{L}}{\\partial{b_2}} = \\delta_3\\\\\n",
"& \\frac{\\partial{L}}{\\partial{W_1}} = x^T \\delta2\\\\\n",
"& \\frac{\\partial{L}}{\\partial{b_1}} = \\delta2 \\\\\n",
"& \\frac{\\partial{L}}{\\partial{W_1}} = x^T \\delta_2\\\\\n",
"& \\frac{\\partial{L}}{\\partial{b_1}} = \\delta_2 \\\\\n",
"\\end{aligned}\n",
"$$"
]
@ -601,4 +601,4 @@
},
"nbformat": 4,
"nbformat_minor": 0
}
}