fix issues

model/neural_network.py

@@ -94,10 +94,9 @@ def backward_pass(self,y,y_pred,l2,lr,**kwargs):
 			grad_wbias = np.sum(grad_wrt_input,axis=0) # (1,200)
 
 			grad_w = self.X_acts[i].T @ grad_wrt_input # (400,200)
-			grad_w+= (l2/bs) * self.weights[i][0] # l2 reg
 
 			grad_w,grad_wbias = self.opt.step(grad_w,grad_wbias,i,**kwargs)
-			self.weights[i][0]-= lr*grad_w #update weight
+			self.weights[i][0]-= lr*(grad_w + (l2/bs) * self.weights[i][0]) #update weight
 			self.weights[i][1]-= lr*grad_wbias #update bias
 
 	def fit_epoch(self,X_train,y_train,X_val,y_val,lr,epochs,bs=64,l2=0,beta1=0.9,beta2=0.99):

model/random_forest.py

@@ -13,7 +13,7 @@ def __init__(self, X, y, n_trees, sample_sz,is_reg=True, min_leaf=3,max_features
 		self.is_reg = is_reg
 
 	def create_tree(self,is_reg,min_leaf,max_features):
-		# generate random idxs with size sample_sz. Thus this model is not suitable for time-related dataset
+		# generate random idxs with size sample_sz
 		sample_idxs = np.random.permutation(len(self.y))[:self.sample_sz]
 		return DecisionTreeNode(self.X[sample_idxs,:], self.y[sample_idxs], is_reg,min_leaf,max_features)
 	def predict(self, X,thres=0.5):

neural_net_optimizers.ipynb

@@ -977,7 +977,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.6"
+   "version": "3.7.4"
   }
  },
  "nbformat": 4,