sar-gupta / neural-network-from-scratch Goto Github PK

Add inverted dropout implementation.

During long waits of training and saving, I was getting confused if the program is stuck. Perhaps progress like the following would help? I can create separate PR or commit to #3 if you think its a good idea.

Starting...
Training...
== EPOCH: 0 / 5 ==
Training with 60000 / 60000
Error: 9.15849543933
== EPOCH: 1 / 5 ==
Training with 23692 / 60000
...
Saving...
Testing...
Accurracy: 7.87

Class names are usually CamelCased, and Python code should follow PEP to maximize readability. The importance of readability is to ensure fellow programmers can better learn from code and track bugs. Nonetheless, your code is generally very readable. Good job! 😊

== EPOCH: 1 / 5 ==
Training with 2001 / 60000
/Users/dikshant/opt/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:101: RuntimeWarning: divide by zero encountered in log
/Users/dikshant/opt/anaconda3/lib/python3.7/site-packages/ipykernel_launcher.py:101: RuntimeWarning: invalid value encountered in multiply
Training with 59951 / 60000/ 60000
Error: nan
== EPOCH: 2 / 5 ==
Training with 59951 / 60000 / 6000060000
Error: nan
== EPOCH: 3 / 5 ==
Training with 59951 / 60000
Error: nan
== EPOCH: 4 / 5 ==
Training with 59951 / 600001551 / 60000
Error: nan
== EPOCH: 5 / 5 ==
Training with 59951 / 60000
Error: nan

this is the output on a run on tensorflow's version of mnist