Jensen-holm commited on
Commit
88a3c01
·
1 Parent(s): ba1e2db

fixed dimensions issue with the iris dataset

Browse files
neural_network/backprop.py CHANGED
@@ -48,8 +48,9 @@ def bp(
48
  error * model.func_prime(y_hat),
49
  )
50
  db2 = np.sum(error * model.func_prime(y_hat), axis=0)
51
- db1 = np.sum(np.dot(error * model.func_prime(y_hat), model.w2.T)
52
- * model.func_prime(node1), axis=0)
 
53
 
54
  # update weights & biases using gradient descent.
55
  # this is -= and not += because if the gradient descent
 
48
  error * model.func_prime(y_hat),
49
  )
50
  db2 = np.sum(error * model.func_prime(y_hat), axis=0)
51
+ db1 = np.sum(
52
+ np.dot(error * model.func_prime(y_hat), model.w2.T) * model.func_prime(node1), axis=0,
53
+ )
54
 
55
  # update weights & biases using gradient descent.
56
  # this is -= and not += because if the gradient descent
neural_network/main.py CHANGED
@@ -5,10 +5,7 @@ from neural_network.opts import activation
5
  from neural_network.backprop import bp
6
 
7
 
8
- def init(
9
- X: np.array,
10
- hidden_size: int
11
- ) -> dict:
12
  """
13
  returns a dictionary containing randomly initialized
14
  weights and biases to start off the neural_network
@@ -16,8 +13,8 @@ def init(
16
  return {
17
  "w1": np.random.randn(X.shape[1], hidden_size),
18
  "b1": np.zeros((1, hidden_size)),
19
- "w2": np.random.randn(hidden_size, 1),
20
- "b2": np.zeros((1, 1)),
21
  }
22
 
23
 
 
5
  from neural_network.backprop import bp
6
 
7
 
8
+ def init(X: np.array, hidden_size: int) -> dict:
 
 
 
9
  """
10
  returns a dictionary containing randomly initialized
11
  weights and biases to start off the neural_network
 
13
  return {
14
  "w1": np.random.randn(X.shape[1], hidden_size),
15
  "b1": np.zeros((1, hidden_size)),
16
+ "w2": np.random.randn(hidden_size, 3), # Output layer has 3 neurons
17
+ "b2": np.zeros((1, 3)), # Output layer has 3 neurons
18
  }
19
 
20