believe to be performing backprop with the weights without error

neural_network/activation.py

@@ -1,23 +1,15 @@
 import numpy as np
 
 
-
 def sigmoid(x: float) -> float:
     return 1.0 / (1.0 + np.exp(-x))
 
 def sigmoid_prime(x: float) -> float:
     return sigmoid(x) / (1.0 - sigmoid())
 
-def relu(x: float) -> float:
-    """
-    returns the input if > 0
-    """
-    return max(0.0, x)
+def relu(x):
+    return np.maximum(x, 0)
 
-def relu_prime(x: float) -> float:
-    """
-    returns 1 if input is +
-    returns 0 if input is -
-    """
-    return 1 if x > 0 else 0
+def relu_prime(x):
+    return np.where(x > 0, 1, 0)
 

neural_network/backprop.py

@@ -1,7 +1,40 @@
 import numpy as np
 
+from neural_network.opts import activation
 
-def bp():
-    return
+
+def bp(X_train: np.array, y_train: np.array, wb: dict, args: dict):
+    epochs = args["epochs"]
+    func = activation[args["activation_func"]]["main"]
+    func_prime = activation[args["activation_func"]]["prime"]
+    w1, w2 = wb["W1"], wb["W2"]
+    b1, b2 = wb["b1"], wb["b2"]
+    lr = args["learning_rate"]
+
+    for e in range(epochs):
+        # forward prop
+        node1 = compute_node(X_train, w1, b1, func)
+        y_hat = compute_node(node1, w2, b2, func)
+        error = y_hat - y_train
+
+        # backprop
+        # right now this is just the weights,
+        # we should also update the biases
+        dw2 = np.dot(
+            node1.T, 
+            error * func_prime(y_hat),
+        )
+        dw1 = np.dot(
+            X_train.T, 
+            np.dot(error * func_prime(y_hat), w2.T) * func_prime(node1),
+        )
+
+        # update weights & biases
+        w1 -= lr * dw1
+        w2 -= lr * dw2
+    
+
+def compute_node(X, w, b, func):
+    return func(np.dot(X, w) + b)
 
 

neural_network/forwardprop.py

@@ -1,6 +0,0 @@
-import numpy as np
-
-
-def fp():
-    return
-

neural_network/main.py

@@ -1,7 +1,6 @@
-import numpy as np
-
-from neural_network.forwardprop import fp
+from sklearn.model_selection import train_test_split
 from neural_network.backprop import bp
+import numpy as np
 
 
 def get_args() -> dict:
@@ -35,13 +34,16 @@ def main(
     X: np.array,
     y: np.array,
 ) -> None:
+
     args = get_args()
     wb = init(X, y, args["hidden_size"])
+    X_train, X_test, y_train, y_test = train_test_split(
+        X, 
+        y, 
+        test_size=0.3, 
+        random_state=8675309
+    )
 
-    for e in range(args["epochs"]):
-        fp()
-        bp()
+    results = bp(X_train, y_train, wb, args)
 
-        # update weights and biases
 
-    # print results

requirements.txt

@@ -1 +1,2 @@
 numpy==1.24.2
+scikit_learn==1.2.2