Refactoring FastXML -> Trainer, Inferencer

fastxml/__init__.py

@@ -1 +1 @@
-from .fastxml import FastXML
+from .fastxml import Trainer, Inferencer

fastxml/fastxml.py

@@ -93,7 +93,51 @@ def dense_rows_iter(dense):
 
  yield ' '.join(dense_lines)
 
-class FastXML(object):
+class Inferencer(object):
+ """
+ Loads up a model for inferencing
+ """
+ def __init__(self, dname, gamma=30, blend=0.8, leaf_probs=False):
+ with file(os.path.join(dname, 'settings')) as f:
+ self.__dict__.update(json.load(f))
+
+ self.gamma = gamma
+ self.blend = blend
+ self.leaf_probs = leaf_probs
+
+ forest = IForest(dname, self.n_trees, self.n_labels)
+ if self.leaf_classifiers:
+ lc = LeafComputer(dname)
+ predictor = Blender(forest, lc)
+ else:
+ predictor = IForestBlender(forest)
+
+ self.predictor = predictor
+
+ def predict(self, X, fmt='sparse'):
+ assert fmt in ('sparse', 'dict')
+ s = []
+ num = X.shape[0] if isinstance(X, sp.csr_matrix) else len(X)
+ for i in xrange(num):
+ Xi = X[i]
+ mean = self.predictor.predict(Xi.data, Xi.indices, self.blend, self.gamma, self.leaf_probs)
+
+ if fmt == 'sparse':
+ s.append(mean)
+
+ else:
+ od = OrderedDict()
+ for idx in reversed(mean.data.argsort()):
+ od[mean.indices[idx]] = mean.data[idx]
+
+ s.append(od)
+
+ if fmt == 'sparse':
+ return sp.vstack(s)
+
+ return s
+
+class Trainer(object):
 
  def __init__(self, n_trees=1, max_leaf_size=10, max_labels_per_leaf=20,
  re_split=0, n_jobs=1, alpha=1e-4, n_epochs=2, n_updates=100, bias=True, 
@@ -298,7 +342,7 @@ def _save_settings(self, dname):
  import json
  settings = {}
  for k, v in self.__dict__.iteritems():
- if k == 'roots' or k.endswith('_'): 
+ if k == 'roots' or k == 'predictor' or k.endswith('_'): 
  continue
 
  settings[k] = v