import?numpy?as?np

class?BPNNClassifier:
????def?__init__（self?feature_n?hidden_n=10?deep=2?label_n=2?eta=0.1?max_iter=200?activate_func=“tanh“）:
????????self.feature_n?=?feature_n
????????self.hidden_n?=?hidden_n
????????self.deep?=?deep
????????self.label_n?=?label_n
????????self.eta?=?eta
????????self.max_iter?=?max_iter
????????self.weights?=?[]
????????self.gradients?=?list（range（deep））?#?save?the?gradient?of?every?neuron
????????self.values?=?[]?#?save?the?activated?value?of?every?neuron
????????activate_funcs?=?\
????????????{“tanh“:（self.tanh?self.dtanh）?“sigmoid“:（self.sigmoid?self.dsigmoid）}
????????self.activate_func?self.dactivate_func?=?activate_funcs[activate_func]
????????for?d?in?range（deep）:
????????????if?d?==?0:?#?input?layer?to?hidden?layer
????????????????weight?=?np.random.randn（hidden_n?feature_n?+?1）
????????????elif?d?==?self.deep?-?1:?#?hidden?layer?to?output?layer
????????????????label_n?=?1?if?label_n?==?2?else?label_n
????????????????weight?=?np.random.randn（label_n?hidden_n）
????????????else:?#?the?others
????????????????label_n?=?1?if?label_n?==?2?else?label_n
????????????????weight?=?np.random.randn（hidden_n?hidden_n）
????????????self.weights.append（weight）

????def?linear_input（self?deep?X）:
????????weight?=?self.weights[deep]
????????return?X?@?weight.T

????def?activation（self?z?func）:
????????return?func（z）

????def?sigmoid（self?z）:
????????return?1?/?（1?+?np.exp（-z））

????def?dsigmoid（self?h）:
????????return?h?*?（1?-?h）

????def?tanh（self?z）:
????????return?（np.exp（z）?-?np.exp（-z））?/?（np.exp（z）?+?np.exp（-z））

????def?dtanh（self?h）:
????????return?1?-?h?**?2

????def?preproccessing（self?X=None?y=None）:
????????X_y?=?[]
????????if?isinstance（X?np.ndarray）:
????????????X0?=?np.array（[[1]?for?i?in?range（X.shape[0]）]）
????????????X?=?np.hstack（[X0?X]）
????????????X_y.append（X）
????????if?isinstance（y?np.ndarray）:
????????????y?=?self.encoder（y）
????????????X_y.append（y）
????????return?tuple（X_y）

????def?encoder（self?y）:
????????y_new?=?[]
????????if?y.ndim?==?1:?#?encode?y?to?one?hot?code
????????????if?self.label_n?>?2:
????????????????for?yi?in?y:
????????????????????yi_new?=?np.zeros（self.label_n）
????????????????????yi_new[yi]?=?1
????????????????????y_new.append（yi_new）
????????????????y_new?=?np.array（y_new）
????????????else:
????????????????y_new?=?y
????????elif?y.ndim?==?2:?#?encode?y?to?1D?array
????????????if?self.label_n?>?2:
????????????????for?yi?in?y:
????????????????????for?j?in?range（len（yi））:
????????????????????????if?yi[j]?==?1:
????????????????????????????y_new.append（j）
????????????????????????????break
????????????????y_new?=?np.array（y_new）
????????????else:
????????????????y_new?=?y.ravel（）
????????else:
????????????raise?Exception（“argument?value?error:?ndarray?ndim?should?be?1?or?2“）
????????return?y_new

????def?forward_propagation（self?X）:
????????self.values.clear（）
????????value?=?None
????????for?d?in?range（self.deep）:
????????????if?d?==?0:?#?input?lay