fixing sklearn binary classifiers

Author: lucastsutsui

embml/sklearnModels/embml_sklearn_LinearSVC.py

@@ -7,13 +7,18 @@ def process(self, model, opts):
  self.coef = model.coef_.tolist()
  self.intercept = model.intercept_.tolist()
  self.classes = list(map(int, model.classes_.tolist()))
+ self.input_size = len(self.coef[0])
 
  if opts['useFxp']:
  self.coef = [[utils.toFxp(self.coef[_i][_j], opts) \
  for _j in range(len(self.coef[_i]))] \
  for _i in range(len(self.coef))]
  self.intercept = [utils.toFxp(_i, opts)\
  for _i in self.intercept]
+ 
+ if len(self.classes) == 2:
+ self.coef = self.coef[0]
+ self.intercept = self.intercept[0]
 
 def write_output(classifier, opts):
  funcs = '\n'
@@ -24,47 +29,65 @@ def write_output(classifier, opts):
  decType = ("FixedNum" if opts['useFxp'] else "float")
 
  # Classify function
- funcs += utils.write_func_init("int", "classify") + \
- utils.write_dec("int", "indMax", initValue="0", tabs=1) + \
- utils.write_dec(decType, "scores[NUM_CLASSES]", tabs=1)
+ funcs += utils.write_func_init("int", "classify")
+
+ if len(classifier.classes) == 2:
+ funcs += utils.write_dec(decType, "score", tabs=1)
+ else:
+ funcs += utils.write_dec("int", "indMax", initValue="0", tabs=1) + \
+ utils.write_dec(decType, "scores[NUM_CLASSES]", tabs=1)
 
  if opts['C']: 
  funcs += utils.write_dec("int", "i", tabs=1) + \
  utils.write_dec("int", "j", tabs=1)
- 
- funcs += utils.write_for("i = 0", "i < NUM_CLASSES", "i++", tabs=1, inC=opts['C']) + \
- utils.write_attribution("scores[i]", "intercept[i]", tabs=2) + \
- utils.write_for("j = 0", "j < INPUT_SIZE", "j++", tabs=2, inC=opts['C']) + \
- utils.write_attribution("scores[i]", \
- ("fxp_sum(scores[i], fxp_mul(coef[i][j], instance[j]))" \
- if opts['useFxp'] else \
- "(coef[i][j] * instance[j])"), \
- op=('' if opts['useFxp'] else '+'), tabs=3) + \
- utils.write_end(tabs=2) + \
- utils.write_if("scores[i] " + \
- ("<" if len(classifier.classes) == 2 else ">") + \
- " scores[indMax]", tabs=2) + \
- utils.write_attribution("indMax", "i", tabs=3) + \
- utils.write_end(tabs=2) + \
- utils.write_end(tabs=1) + \
- utils.write_ret("classes[indMax]", tabs=1) + \
- utils.write_end(tabs=0)
+
+ if len(classifier.classes) == 2:
+ funcs += utils.write_attribution("score", "intercept", tabs=1) + \
+ utils.write_for("i = 0", "i < INPUT_SIZE", "i++", tabs=1, inC=opts['C']) + \
+ utils.write_attribution("score", \
+ ("fxp_sum(score, fxp_mul(coef[i], instance[i]))" \
+ if opts['useFxp'] else \
+ "(coef[i] * instance[i])"), \
+ op=('' if opts['useFxp'] else '+'), tabs=2) + \
+ utils.write_end(tabs=1) + \
+ utils.write_ret("classes[((score > 0) ? 1 : 0)]", tabs=1) + \
+ utils.write_end(tabs=0)
+ else:
+ funcs += utils.write_for("i = 0", "i < NUM_CLASSES", "i++", tabs=1, inC=opts['C']) + \
+ utils.write_attribution("scores[i]", "intercept[i]", tabs=2) + \
+ utils.write_for("j = 0", "j < INPUT_SIZE", "j++", tabs=2, inC=opts['C']) + \
+ utils.write_attribution("scores[i]", \
+ ("fxp_sum(scores[i], fxp_mul(coef[i][j], instance[j]))" \
+ if opts['useFxp'] else \
+ "(coef[i][j] * instance[j])"), \
+ op=('' if opts['useFxp'] else '+'), tabs=3) + \
+ utils.write_end(tabs=2) + \
+ utils.write_if("scores[i] " + \
+ ("<" if len(classifier.classes) == 2 else ">") + \
+ " scores[indMax]", tabs=2) + \
+ utils.write_attribution("indMax", "i", tabs=3) + \
+ utils.write_end(tabs=2) + \
+ utils.write_end(tabs=1) + \
+ utils.write_ret("classes[indMax]", tabs=1) + \
+ utils.write_end(tabs=0)
 
  # Declaration of global variables
  decls += utils.write_dec(decType, "instance[INPUT_SIZE + 1]") + '\n' + \
  utils.write_dec("const " + decType, \
- "coef[NUM_CLASSES][INPUT_SIZE]", \
+ ("coef[INPUT_SIZE]"
+ if len(classifier.classes) == 2 else
+ "coef[NUM_CLASSES][INPUT_SIZE]"), \
  initValue=utils.toStr(classifier.coef)) + '\n' + \
  utils.write_dec("const " + decType, \
- "intercept[NUM_CLASSES]", \
+ "intercept" + ("[NUM_CLASSES]" if len(classifier.classes) != 2 else ""), \
  initValue=utils.toStr(classifier.intercept)) + '\n' + \
  utils.write_dec("const " + utils.chooseDataType(classifier.classes), \
  "classes[NUM_CLASSES]", \
  initValue=utils.toStr(classifier.classes)) + '\n'
 
  # Definition of constant values
  defs += utils.write_define("NUM_CLASSES", str(len(classifier.classes))) + \
- utils.write_define("INPUT_SIZE", str(len(classifier.coef[0])))
+ utils.write_define("INPUT_SIZE", str(classifier.input_size))
 
  # Include of libraries
  if opts['useFxp']:

embml/sklearnModels/embml_sklearn_LogisticRegression.py

@@ -7,13 +7,18 @@ def process(self, model, opts):
  self.coef = model.coef_.tolist()
  self.intercept = model.intercept_.tolist()
  self.classes = list(map(int, model.classes_.tolist()))
-
+ self.input_size = len(self.coef[0])
+ 
  if opts['useFxp']:
  self.coef = [[utils.toFxp(self.coef[_i][_j], opts) \
  for _j in range(len(self.coef[_i]))] \
  for _i in range(len(self.coef))]
  self.intercept = [utils.toFxp(_i, opts)\
  for _i in self.intercept]
+ 
+ if len(self.classes) == 2:
+ self.coef = self.coef[0]
+ self.intercept = self.intercept[0]
 
 def write_output(classifier, opts):
  funcs = '\n'
@@ -24,47 +29,65 @@ def write_output(classifier, opts):
  decType = ("FixedNum" if opts['useFxp'] else "float")
 
  # Classify function
- funcs += utils.write_func_init("int", "classify") + \
- utils.write_dec("int", "indMax", initValue="0", tabs=1) + \
- utils.write_dec(decType, "scores[NUM_CLASSES]", tabs=1)
+ funcs += utils.write_func_init("int", "classify")
+
+ if len(classifier.classes) == 2:
+ funcs += utils.write_dec(decType, "score", tabs=1)
+ else:
+ funcs += utils.write_dec("int", "indMax", initValue="0", tabs=1) + \
+ utils.write_dec(decType, "scores[NUM_CLASSES]", tabs=1)
 
  if opts['C']: 
  funcs += utils.write_dec("int", "i", tabs=1) + \
  utils.write_dec("int", "j", tabs=1)
- 
- funcs += utils.write_for("i = 0", "i < NUM_CLASSES", "i++", tabs=1, inC=opts['C']) + \
- utils.write_attribution("scores[i]", "intercept[i]", tabs=2) + \
- utils.write_for("j = 0", "j < INPUT_SIZE", "j++", tabs=2, inC=opts['C']) + \
- utils.write_attribution("scores[i]", \
- ("fxp_sum(scores[i], fxp_mul(coef[i][j], instance[j]))" \
- if opts['useFxp'] else \
- "(coef[i][j] * instance[j])"), \
- op=('' if opts['useFxp'] else '+'), tabs=3) + \
- utils.write_end(tabs=2) + \
- utils.write_if("scores[i] " + \
- ("<" if len(classifier.classes) == 2 else ">") + \
- " scores[indMax]", tabs=2) + \
- utils.write_attribution("indMax", "i", tabs=3) + \
- utils.write_end(tabs=2) + \
- utils.write_end(tabs=1) + \
- utils.write_ret("classes[indMax]", tabs=1) + \
- utils.write_end(tabs=0)
+
+ if len(classifier.classes) == 2:
+ funcs += utils.write_attribution("score", "intercept", tabs=1) + \
+ utils.write_for("i = 0", "i < INPUT_SIZE", "i++", tabs=1, inC=opts['C']) + \
+ utils.write_attribution("score", \
+ ("fxp_sum(score, fxp_mul(coef[i], instance[i]))" \
+ if opts['useFxp'] else \
+ "(coef[i] * instance[i])"), \
+ op=('' if opts['useFxp'] else '+'), tabs=2) + \
+ utils.write_end(tabs=1) + \
+ utils.write_ret("classes[((score > 0) ? 1 : 0)]", tabs=1) + \
+ utils.write_end(tabs=0)
+ else:
+ funcs += utils.write_for("i = 0", "i < NUM_CLASSES", "i++", tabs=1, inC=opts['C']) + \
+ utils.write_attribution("scores[i]", "intercept[i]", tabs=2) + \
+ utils.write_for("j = 0", "j < INPUT_SIZE", "j++", tabs=2, inC=opts['C']) + \
+ utils.write_attribution("scores[i]", \
+ ("fxp_sum(scores[i], fxp_mul(coef[i][j], instance[j]))" \
+ if opts['useFxp'] else \
+ "(coef[i][j] * instance[j])"), \
+ op=('' if opts['useFxp'] else '+'), tabs=3) + \
+ utils.write_end(tabs=2) + \
+ utils.write_if("scores[i] " + \
+ ("<" if len(classifier.classes) == 2 else ">") + \
+ " scores[indMax]", tabs=2) + \
+ utils.write_attribution("indMax", "i", tabs=3) + \
+ utils.write_end(tabs=2) + \
+ utils.write_end(tabs=1) + \
+ utils.write_ret("classes[indMax]", tabs=1) + \
+ utils.write_end(tabs=0)
 
  # Declaration of global variables
  decls += utils.write_dec(decType, "instance[INPUT_SIZE + 1]") + '\n' + \
  utils.write_dec("const " + decType, \
- "coef[NUM_CLASSES][INPUT_SIZE]", \
+ ("coef[INPUT_SIZE]"
+ if len(classifier.classes) == 2 else
+ "coef[NUM_CLASSES][INPUT_SIZE]"), \
  initValue=utils.toStr(classifier.coef)) + '\n' + \
  utils.write_dec("const " + decType, \
- "intercept[NUM_CLASSES]", \
+ "intercept" + ("[NUM_CLASSES]" if len(classifier.classes) != 2 else ""), \
  initValue=utils.toStr(classifier.intercept)) + '\n' + \
  utils.write_dec("const " + utils.chooseDataType(classifier.classes), \
  "classes[NUM_CLASSES]", \
  initValue=utils.toStr(classifier.classes)) + '\n'
 
  # Definition of constant values
  defs += utils.write_define("NUM_CLASSES", str(len(classifier.classes))) + \
- utils.write_define("INPUT_SIZE", str(len(classifier.coef[0])))
+ utils.write_define("INPUT_SIZE", str(classifier.input_size))
 
  # Include of libraries
  if opts['useFxp']: