Merge branch 'master' of https://github.com/acmucsd/acm-ai-workshops

Author: edwardyang12

Fall20/NeuralNetworks1/NN.py

@@ -7,6 +7,17 @@ def linear_activation(z):
 def tanh_activation(z):
     return np.tanh(z)
 
+
+def averageOf3(input1, input2, input3):
+    w1 = 1.0 / 3.0
+    w2 = 1.0 / 3.0
+    w3 = 1.0 / 3.0
+    bias = 0
+    z = input1 * w1 + input2 * w2 + input3 * w3 + bias
+    y = linear_activation(z)
+    return y
+
+
 # 2 layer NN for implementation of OR gate
 def orgate(input1, input2):
     bias = -1
@@ -31,3 +42,5 @@ def boolToBinary(bool1,bool2):
 
 input1, input2 = boolToBinary(True,True)
 print(orgate(input1,input2))
+
+print(averageOf3(1, 0, 3))

Fall20/NeuralNetworks1/README.md

@@ -1,14 +1,15 @@
 # 1. Introduction to AI and Neural Networks
 
-Welcome to the repo for ACM AI's first workshop on AI and Neural Networks, hosted in fall 2020
+Welcome to the repo for ACM AI's first workshop on AI and Neural Networks, hosted in fall 2020. Slides are here:
+https://docs.google.com/presentation/d/1a9TwuZkqAwEvo1eCFcL-Hub1OiJNxzidvtBaL3OwLWQ/edit?usp=sharing
 
 Here's a quick run down of what was in the workshop!
 
 1. What is AI?
 2. Stats and history of AI
 3. What is AI all about? Is it just Neural Networks? (answer is no)
 4. Applications
-5. What is a Neural Network? 
+5. What is a Neural Network?
 6. What is a Neuron? Biologically and Computationally?
 7. Weights and Biases of a Network
 8. Activation Functions
@@ -17,6 +18,7 @@ Here's a quick run down of what was in the workshop!
 11. Resources
 
 In this folder there is also code for various work we demoed in the workshop. We use the `numpy` package to do our demos. To download it, do
+
 ```
 pip install numpy
-```
+```

Fall20/NeuralNetworks1/vectorized.py

@@ -59,24 +59,34 @@ def random_nn(x):
 
     return a_3
 
-print("On 3 layer network, input {} fed forward gives {}".format(x, random_nn(x)))
+# print("On 3 layer network, input {} fed forward gives {}".format(x, random_nn(x)))
 
 # 4 layer NN for computing whether absolute difference is between 1 and 3
 # if between 1 and 3 outputs >0 else output <=0
 def multilayer(x):
+
+    # layer 2
     w1 = np.array([1,-1])
     b1 = 0
     weighted_input1 = np.matmul(w1,x) + b1
-    output1 = parametric_activation(-1,weighted_input1)
+
+    # output of layer 2
+    output2 = parametric_activation(-1, weighted_input1)
+
+    # layer 3
     w2 = np.array([1])
     b2 = -2
-    weighted_input2 = np.matmul(w2,[output1]) + b2
-    output2 = parametric_activation(-1,weighted_input2)
+    weighted_input2 = np.matmul(w2, [output2]) + b2
+
+    # output of layer 3
+    output3 = parametric_activation(-1, weighted_input2)
+
+    # final layer!
     w3 = np.array([-1])
     b3 = 1
-    weighted_input3 = np.matmul(w3,[output2]) + b3
+    weighted_input3 = np.matmul(w3, [output3]) + b3
     y = tanh_activation(weighted_input3)
     return y
 
 x = np.array([4,5.5])
-print(multilayer(x))
+# print(multilayer(x))

Fall20/NeuralNetworks1/worksheet/averagenn.py

@@ -0,0 +1,24 @@
+import numpy as np
+
+# fill in this neural network and activation function to calculate the of average of 5 values using numpy and vectorization
+def activation(z):
+    raise NotImplementedError
+
+def average5NN(x):
+    # this should be a simple 2 layer network, one input layer and one output layer, figure out how many neurons are in each layer, the weights, and biases! 
+
+    # This should take input column vector x and apply an appropriate weight matrix and add a bias, apply the activation and then return the result
+
+    weights = np.array([]) # fill me in!
+
+    bias = 100 # fix me!
+
+    weighted_input = 0 # fill me in!
+
+    raise NotImplementedError
+
+
+# Your network works if you don't get any errors!
+assert average5NN(np.array([1, 1, 1, 1, 1])) == 1
+assert average5NN(np.array([1, -1, 1, -1, 0])) == 0
+assert average5NN(np.array([100, 200, 300, 400, -400])) == 1.2

Fall20/NeuralNetworks1/worksheet/squarenn.py

@@ -0,0 +1,15 @@
+# fill in this neural network and activation function to calculate the square of the sum of 2 inputs
+def activation(z):
+    raise NotImplementedError
+
+def squareNN(x1, x2):
+    # this should be a simple 2 layer network, one input layer and one output layer, figure out how many neurons are in each layer, the weights, and biases!
+    # This should take x1 and x2 and return the square of the sum of the 2 inputs, but obviously in neural network style!
+    raise NotImplementedError
+
+
+# Your network works if you don't get any errors!
+assert squareNN(1, 1) == 4
+assert squareNN(-1, 1) == 0
+assert squareNN(10, 2.5) == 156.25
+assert squareNN(0, 0) == 0

README.md

@@ -2,4 +2,4 @@
 
 This is the repository with all workshop content code etc. for ACM AI, with content organized by quarter offered
 
-Check out https://ai.acmucsd.com for more details about upcoming workshops and the exciting stuff going on at ACM AI!
+Check out https://ai.acmucsd.com for more details about upcoming workshops and the exciting stuff going on at ACM AI!