[impl] Implement MutilSVR

Author: nwtgck

src/multi_svr.py

@@ -1,2 +1,76 @@
-def hello():
-  return "hello, world from multi_svr"
+import numpy as np
+import sklearn
+from sklearn import svm
+
+
+class MutilSVR(sklearn.base.BaseEstimator, sklearn.base.RegressorMixin):
+  def __init__(self, **kwargs):
+    self.__init_kwargs = kwargs
+
+  def fit(self, X, y, **kwargs):
+    X = np.array(X)
+    y = np.array(y)
+
+    # Get dimension of y
+    y_dim = np.ndim(y)
+    if(y_dim == 2):
+      # Feature dimension
+      feature_dim = len(y[0])
+      # Create SVRs
+      self.svrs   = [svm.SVR(**self.__init_kwargs) for _ in range(feature_dim)]
+
+      # For each SVR
+      for curr_feature_dim, svr in enumerate(self.svrs): # (curr=Current)
+        # Select y
+        selected_y = y[:,curr_feature_dim]
+        # Fit
+        svr.fit(X, selected_y, **kwargs)
+    else:
+      raise Exception("Dimension of y must be 2, but found %d" % y_dim)
+
+
+  def predict(self, X):
+    # Init predict list
+    preds = []
+    # For each SVR
+    for curr_feature_dim, svr in enumerate(self.svrs):  # (curr=Current)
+      # Predict
+      pred = svr.predict(X)
+      # Append to preds
+      preds.append(pred)
+
+    pred = np.column_stack(tuple(preds))
+    return pred
+
+
+
+
+if __name__ == '__main__':
+  from sklearn import metrics
+  X = [
+    [0,  0],
+    [0, 10],
+    [1, 10],
+    [1, 20],
+    [1, 30],
+    [1, 40]
+  ]
+
+  y = [
+    [0,  0],
+    [0, 10],
+    [2, 10],
+    [2, 20],
+    [2, 30],
+    [2, 40]
+  ]
+
+  regressor = MutilSVR(kernel='linear')
+
+  regressor.fit(X, y)
+
+  pred_y = regressor.predict(X)
+  err    = metrics.mean_squared_error(y, pred_y, multioutput='raw_values')
+
+  print('pred_y:', pred_y)
+  print('err:', err)