Browse Source

ml notes

master
Markus Bergholz 3 years ago
parent
commit
176948b0a4
  1. 14
      til.md

14
til.md

@ -331,3 +331,17 @@ def dtconverter(o):
print(json.dumps(my_py_dict_var, default = dtconverter))
```
## ML
>> Neural networks are universal approximators - meaning that for any function F and error E, there exists some neural network (needing only a single hidden layer)
that can approximate F with error less than E.
* https://en.wikipedia.org/wiki/Artificial_neural_network#Theoretical_properties
* https://en.wikipedia.org/wiki/Multilayer_perceptron
>> Normalisation is required so that all the inputs are at a comparable range.
With two inputs (x1 and x2), where x1 values are from range 0 to 0.5 and x2 values are from range to 0 to 1000. When x1 is changing by 0.5, the change is 100%, and a
change of x2 by 0.5 is only 0.05%.
Loading…
Cancel
Save