Common functions
Winner searching
These functions find the winning codebook entrys (knn nearest
neighbours of sample) in codebook codes using euclidean
distance. Information about the winning entry is saved in the
winner_info structures (win) provided by the caller. Win
should be an array of knn winner_info structures. When
successful the functions return the number of neighbours found which
should be the same as knn and 0 when winner could not be found
(for example, all components of data vector have been masked off).
- int find_winner_euc(struct entries *codes, struct data_entry *sample, struct winner_info *win, int knn)
- Finds the winning entry (nearest neighbour of
sample). This function only finds one best matching unit
and should be called with knn as one.
- int find_winner_knn(struct entries *codes, struct data_entry *sample, struct winner_info *win, int knn)
- Finds the winning entrys (knn nearest neighbours of
sample).
Vector operations
- float vector_dist_euc(struct data_entry *v1, struct data_entry *v2, int dim)
- Computes the distance between two vectors v1and v2
in euclidean metric. Returns < 0 if distance couldn't be
calculated (all components were masked off). Dim is the
dimension of the vectors. Note that the winner searching
functions don't use this function.
- void adapt_vector(struct data_entry *codetmp, struct data_entry *sample, int dim, float alpha)
- Adapts a codebook vector codetmp by moving it towards the
sample vector by a small amount
(alpha). Dim is the dimension of the vectors.
Alpha functions
These functions return the alpha value for some number of iterations
(iter) out of the maximum number of iterations
(length). Alpha is the initial value when iter is
0.
- float linear_alpha(long iter, long length, float alpha)
- Linearly decreasing alpha.
- float inverse_t_alpha(long iter, long length, float alpha)
- Alpha that decreases proportional to the inverse of iterations
done so far.
Other functions
- int global_options(int argc, char **argv)
- Handles some options that are common to all programs. Also read
some environment variables. Takes care of the options
-compress_cmd, -uncompress_cmd, -mask_str,
-version, -v and environment variables
LVQSOM_COMPRESS_COMMAND,
LVQSOM_UNCOMPRESS_COMMAND, and
LVQSOM_MASK_STR. See the package document or the
new features page for more details
about them. This function should be called in the start of all
programs. Argc and argv are the arguments that
were passed to the main function.
SOM/LVQ Programming info /
SOM/LVQ Home Page /
Neural Networks Research Centre