mirror of https://github.com/glouw/tinn
doc update
This commit is contained in:
parent
63ba88bbb8
commit
10eb871ca7
20
README.md
20
README.md
|
@ -84,6 +84,26 @@ This gives 10 outputs to the neural network. The test program will output the
|
|||
accuracy for each digit. Expect above 99% accuracy for the correct digit, and
|
||||
less that 1% accuracy for the other digits.
|
||||
|
||||
# Tips
|
||||
|
||||
* Tinn will never use more than the C standard library.
|
||||
|
||||
* Tinn is great for embedded systems. Train a model on your powerful desktop and load
|
||||
it onto a microcontroller and use the analog to digital converter to predict real time events.
|
||||
|
||||
* The Tinn source code will always be less than 200 lines. Functions externed in the Tinn header
|
||||
are protected with the _xt_ namespace standing for _externed tinn_.
|
||||
|
||||
* Tinn can easily be multi-threaded with a bit of ingenuity but the master branch will remain
|
||||
single threaded to aid development for embedded systems.
|
||||
|
||||
* Tinn does not seed the random number generator. Do not forget to do so yourself.
|
||||
|
||||
* Always shuffle your input data. Shuffle again after every training iteration.
|
||||
|
||||
* Get greater training accuracy by annealing your learning rate. For instance, multiply
|
||||
your learning rate by 0.99 every training iteration. This will zero in on a good learning minima.
|
||||
|
||||
# Disclaimer
|
||||
|
||||
Tinn is not a fully featured neural network C library like Kann, or Genann:
|
||||
|
|
9
Tinn.c
9
Tinn.c
|
@ -121,12 +121,15 @@ static void* ecalloc(const size_t nmemb, const size_t size)
|
|||
return mem;
|
||||
}
|
||||
|
||||
// Returns an output prediction given an input.
|
||||
float* xtpredict(const Tinn t, const float* const in)
|
||||
{
|
||||
fprop(t, in);
|
||||
return t.o;
|
||||
}
|
||||
|
||||
// Trains a tinn with an input and target output with a learning rate.
|
||||
// Returns error rate of the neural network.
|
||||
float xttrain(const Tinn t, const float* const in, const float* const tg, float rate)
|
||||
{
|
||||
fprop(t, in);
|
||||
|
@ -134,6 +137,9 @@ float xttrain(const Tinn t, const float* const in, const float* const tg, float
|
|||
return toterr(tg, t.o, t.nops);
|
||||
}
|
||||
|
||||
// Builds a new tinn object given number of inputs (nips),
|
||||
// number of hidden neurons for the hidden layer (nhid),
|
||||
// and number of outputs (nops).
|
||||
Tinn xtbuild(const int nips, const int nhid, const int nops)
|
||||
{
|
||||
Tinn t;
|
||||
|
@ -152,6 +158,7 @@ Tinn xtbuild(const int nips, const int nhid, const int nops)
|
|||
return t;
|
||||
}
|
||||
|
||||
// Saves the tinn to disk.
|
||||
void xtsave(const Tinn t, const char* const path)
|
||||
{
|
||||
FILE* const file = efopen(path, "w");
|
||||
|
@ -163,6 +170,7 @@ void xtsave(const Tinn t, const char* const path)
|
|||
fclose(file);
|
||||
}
|
||||
|
||||
// Loads a new tinn from disk.
|
||||
Tinn xtload(const char* const path)
|
||||
{
|
||||
FILE* const file = efopen(path, "r");
|
||||
|
@ -180,6 +188,7 @@ Tinn xtload(const char* const path)
|
|||
return t;
|
||||
}
|
||||
|
||||
// Frees a tinn from the heap.
|
||||
void xtfree(const Tinn t)
|
||||
{
|
||||
free(t.w);
|
||||
|
|
47
Tinn.h
47
Tinn.h
|
@ -2,38 +2,39 @@
|
|||
|
||||
typedef struct
|
||||
{
|
||||
float* w; // All the weights.
|
||||
float* x; // Hidden to output layer weights.
|
||||
float* b; // Biases.
|
||||
float* h; // Hidden layer.
|
||||
float* o; // Output layer.
|
||||
// All the weights.
|
||||
float* w;
|
||||
// Hidden to output layer weights.
|
||||
float* x;
|
||||
// Biases.
|
||||
float* b;
|
||||
// Hidden layer.
|
||||
float* h;
|
||||
// Output layer.
|
||||
float* o;
|
||||
|
||||
int nb; // Number of biases - always two - Tinn only supports a single hidden layer.
|
||||
int nw; // Number of weights.
|
||||
// Number of biases - always two - Tinn only supports a single hidden layer.
|
||||
int nb;
|
||||
// Number of weights.
|
||||
int nw;
|
||||
|
||||
int nips; // Number of inputs.
|
||||
int nhid; // Number of hidden neurons.
|
||||
int nops; // Number of outputs.
|
||||
// Number of inputs.
|
||||
int nips;
|
||||
// Number of hidden neurons.
|
||||
int nhid;
|
||||
// Number of outputs.
|
||||
int nops;
|
||||
}
|
||||
Tinn;
|
||||
|
||||
// Trains a tinn with an input and target output with a learning rate.
|
||||
// Returns error rate of the neural network.
|
||||
float xttrain(Tinn, const float* in, const float* tg, float rate);
|
||||
|
||||
// Builds a new tinn object given number of inputs (nips),
|
||||
// number of hidden neurons for the hidden layer (nhid),
|
||||
// and number of outputs (nops).
|
||||
Tinn xtbuild(int nips, int nhid, int nops);
|
||||
|
||||
// Returns an output prediction given an input.
|
||||
float* xtpredict(Tinn, const float* in);
|
||||
|
||||
// Saves the tinn to disk.
|
||||
float xttrain(Tinn, const float* in, const float* tg, float rate);
|
||||
|
||||
Tinn xtbuild(int nips, int nhid, int nops);
|
||||
|
||||
void xtsave(Tinn, const char* path);
|
||||
|
||||
// Loads a new tinn from disk.
|
||||
Tinn xtload(const char* path);
|
||||
|
||||
// Frees a tinn from the heap.
|
||||
void xtfree(Tinn);
|
||||
|
|
Loading…
Reference in New Issue