added gridsearch and extended gensvm class

2 files changed, 151 insertions, 6 deletions

diff --git a/src/wrapper.pxd b/src/wrapper.pxd
index be4d5f5..441c15b 100644
--- a/src/wrapper.pxd
+++ b/src/wrapper.pxd
@@ -21,7 +21,6 @@ cdef extern from "gensvm_sparse.h":
     GenSparse *gensvm_init_sparse()
     void gensvm_free_sparse(GenSparse *)
 
-
 cdef extern from "gensvm_base.h":
 
     cdef struct GenData:
@@ -64,6 +63,33 @@ cdef extern from "gensvm_base.h":
     GenData *gensvm_init_data()
     void gensvm_free_data(GenData *)
 
+
+cdef extern from "gensvm_task.h":
+
+    cdef struct GenTask:
+        long ID
+        long folds
+        GenData *train_data
+        GenData *test_data
+
+        KernelType kerneltype
+        int weight_idx
+        double p
+        double kappa
+        double lmd
+        double epsilon
+        double gamma
+        double coef
+        double degree
+        double max_iter
+
+        double performance
+        double duration
+        long *predictions
+
+    GenTask *gensvm_init_task()
+    gensvm_free_task(GenTask *)
+
 cdef extern from "gensvm_train.h":
 
     void gensvm_train(GenModel *, GenData *, GenModel *) nogil
@@ -72,12 +98,26 @@ cdef extern from "gensvm_sv.h":
 
     long gensvm_num_sv(GenModel *)
 
+cdef extern from "gensvm_queue.h":
+
+    cdef struct GenQueue:
+        GenTask **tasks
+        long N
+        long i
+
+    GenQueue *gensvm_init_queue()
+    void gensvm_free_queue(GenQueue *)
+
 cdef extern from "gensvm_helper.c":
 
     ctypedef char* char_const_ptr "char const *"
     void set_model(GenModel *, double, double, double, double, int, int, 
             double, double, double, double, long, long)
+    void set_seed_model(GenModel *, double, double, double, double, int, int, 
+            double, double, double, double, long, long, char *, long, long)
     void set_data(GenData *, char *, char *, np.npy_intp *, long)
+    void set_task(GenTask *, int, GenData *, int, double, double, double, 
+            double, double, int, double, double, double, long)
     char_const_ptr check_model(GenModel *)
     void copy_V(void *, GenModel *)
     long get_iter_count(GenModel *)
@@ -89,3 +129,8 @@ cdef extern from "gensvm_helper.c":
     void free_data(GenData *)
     void set_verbosity(int)
     void gensvm_predict(char *, char *, long, long, long, char *) nogil
+    void gensvm_train_q_helper(GenQueue *, char *, int) nogil
+    void set_queue(GenQueue *, long, GenTask **)
+    double get_task_duration(GenTask *)
+    double get_task_performance(GenTask *)
+    void copy_task_predictions(GenTask *, char *, long)
diff --git a/src/wrapper.pyx b/src/wrapper.pyx
index 1d84b59..4ded637 100644
--- a/src/wrapper.pyx
+++ b/src/wrapper.pyx
@@ -11,6 +11,8 @@ Not implemented yet:
 
 from __future__ import print_function
 
+from libc.stdlib cimport malloc, free
+
 import numpy as np
 cimport numpy as np
 
@@ -23,6 +25,7 @@ GENSVM_KERNEL_TYPES = ["linear", "poly", "rbf", "sigmoid"]
 def train_wrap(
         np.ndarray[np.float64_t, ndim=2, mode='c'] X,
         np.ndarray[np.int_t, ndim=1, mode='c'] y,
+        long n_class,
         double p=1.0,
         double lmd=pow(2, -8),
         double kappa=0.0,
@@ -34,25 +37,25 @@ def train_wrap(
         double degree=2.0,
         double kernel_eigen_cutoff=1e-8,
         int max_iter=100000000,
-        int random_seed=-1):
+        int random_seed=-1,
+        np.ndarray[np.float64_t, ndim=2, mode='c'] seed_V=None
+        ):
     """
     """
 
     # Initialize model and data
     cdef GenModel *model = gensvm_init_model()
     cdef GenData *data = gensvm_init_data()
+    cdef GenModel *seed_model = gensvm_init_model()
     cdef long n_obs
     cdef long n_var
-    cdef long n_class
 
     # get the kernel index
     kernel_index = GENSVM_KERNEL_TYPES.index(kernel)
 
     # get the number of classes
-    classes = np.unique(y)
     n_obs = X.shape[0]
     n_var = X.shape[1]
-    n_class = classes.shape[0]
 
     # Set the data
     set_data(data, X.data, y.data, X.shape, n_class)
@@ -61,6 +64,14 @@ def train_wrap(
     set_model(model, p, lmd, kappa, epsilon, weight_idx, kernel_index, degree, 
             gamma, coef, kernel_eigen_cutoff, max_iter, random_seed)
 
+    if not seed_V is None:
+        set_seed_model(seed_model, p, lmd, kappa, epsilon, weight_idx, 
+                kernel_index, degree, gamma, coef, kernel_eigen_cutoff, 
+                max_iter, random_seed, seed_V.data, n_var, n_class)
+    else:
+        gensvm_free_model(seed_model)
+        seed_model = NULL
+
     # Check the parameters
     error_msg = check_model(model)
     if error_msg:
@@ -71,7 +82,10 @@ def train_wrap(
 
     # Do the actual training
     with nogil:
-        gensvm_train(model, data, NULL)
+        gensvm_train(model, data, seed_model)
+
+    # update the number of variables (this may have changed due to kernel)
+    n_var = get_m(model)
 
     # copy the results
     cdef np.ndarray[np.float64_t, ndim=2, mode='c'] V
@@ -90,6 +104,7 @@ def train_wrap(
 
     return (V, n_SV, iter_count, training_error, fit_status)
 
+
 def predict_wrap(
         np.ndarray[np.float64_t, ndim=2, mode='c'] X,
         np.ndarray[np.float64_t, ndim=2, mode='c'] V
@@ -116,6 +131,91 @@ def predict_wrap(
 
     return predictions
 
+def grid_wrap(
+        np.ndarray[np.float64_t, ndim=2, mode='c'] X,
+        np.ndarray[np.int_t, ndim=1, mode='c'] y,
+        candidate_params,
+        int store_predictions,
+        np.ndarray[np.int_t, ndim=1, mode='c'] cv_idx,
+        int n_folds,
+        ):
+    """
+    """
+
+    cdef GenQueue *queue = gensvm_init_queue()
+    cdef GenData *data = gensvm_init_data()
+    cdef GenTask *task
+    cdef long n_obs
+    cdef long n_var
+    cdef long n_class
+    cdef long n_tasks = len(candidate_params)
+
+    # get the number of classes
+    classes = np.unique(y)
+    n_obs = X.shape[0]
+    n_var = X.shape[1]
+    n_class = classes.shape[0]
+
+    set_data(data, X.data, y.data, X.shape, n_class)
+
+    cdef GenTask **tasks = <GenTask **>malloc(n_tasks * sizeof(GenTask *))
+
+    ID = 0
+    for candidate in candidate_params:
+        params = {
+                'p': 1.0,
+                'lmd': 1e-5,
+                'kappa': 0.0,
+                'epsilon': 1e-6,
+                'weight_idx': 1,
+                'kernel': GENSVM_KERNEL_TYPES.index('linear'),
+                'gamma': 1.0,
+                'coef': 0.0,
+                'degree': 2.0,
+                'max_iter': 1e8
+                }
+        params.update(candidate)
+        if 'kernel' in candidate:
+            params['kernel'] = GENSVM_KERNEL_TYPES.index(candidate['kernel'])
+        if 'weights' in candidate:
+            params['weight_idx'] = 1 if candidate['weights'] == 'unit' else 2
+
+        task = gensvm_init_task()
+        set_task(task, ID, data, n_folds, params['p'], params['lmd'], 
+                params['kappa'], params['epsilon'], params['weight_idx'], 
+                params['kernel'], params['degree'], params['gamma'], 
+                params['coef'], params['max_iter'])
+
+        tasks[ID] = task
+        ID += 1
+
+    set_queue(queue, n_tasks, tasks)
+
+    with nogil:
+        gensvm_train_q_helper(queue, cv_idx.data, store_predictions)
+
+    cdef np.ndarray[np.int_t, ndim=1, mode='c'] pred
+
+    results = dict()
+    results['params'] = []
+    results['duration'] = []
+    results['scores'] = []
+    results['predictions'] = []
+    for ID in range(n_tasks):
+        results['params'].append(candidate_params[ID])
+        results['duration'].append(get_task_duration(tasks[ID]))
+        results['scores'].append(get_task_performance(tasks[ID]))
+        if store_predictions:
+            pred = np.zeros((n_obs, ), dtype=np.int)
+            copy_task_predictions(tasks[ID], pred.data, n_obs)
+            results['predictions'].append(pred.copy())
+
+    gensvm_free_queue(queue)
+    free_data(data)
+
+    return results
+
+
 def set_verbosity_wrap(int verbosity):
     """
     Control verbosity of gensvm library