From 662ab4d81813e76fc039e64eb6ba2a0f88dc625b Mon Sep 17 00:00:00 2001
From: Claudio Scheer <claudioscheer@protonmail.com>
Date: Tue, 31 Mar 2020 02:23:12 -0300
Subject: [PATCH] Perceptron neural network to handle OR operator

---
 perceptron-or/.gitignore |  1 +
 perceptron-or/README.md  |  3 +++
 perceptron-or/main.py    | 41 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 45 insertions(+)
 create mode 100644 perceptron-or/.gitignore
 create mode 100644 perceptron-or/README.md
 create mode 100644 perceptron-or/main.py

diff --git a/perceptron-or/.gitignore b/perceptron-or/.gitignore
new file mode 100644
index 0000000..722d5e7
--- /dev/null
+++ b/perceptron-or/.gitignore
@@ -0,0 +1 @@
+.vscode
diff --git a/perceptron-or/README.md b/perceptron-or/README.md
new file mode 100644
index 0000000..ef4e0cb
--- /dev/null
+++ b/perceptron-or/README.md
@@ -0,0 +1,3 @@
+# About
+
+This is an example of a perceptron neural network that can learn how to deal with the OR operator. This neural network cannot handle XOR operator, because XOR cannot be expressed as a linear function like OR.
\ No newline at end of file
diff --git a/perceptron-or/main.py b/perceptron-or/main.py
new file mode 100644
index 0000000..b852d0d
--- /dev/null
+++ b/perceptron-or/main.py
@@ -0,0 +1,41 @@
+import numpy as np
+
+# np.random.seed(0)
+
+train_data = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
+train_data_y = np.array([-1, 1, 1, 1])
+
+
+def threshold_activation_function(x):
+    return -1 if x < 0 else 1
+
+
+# Add the bias to the input data.
+# size = 4x3
+train_data = np.insert(train_data, 0, 1, axis=1)
+
+epochs = 1000
+random_weight_threshold = 0.5
+learning_rate = 0.001
+
+# Synapses that will be learned.
+# size = 3x1
+weigths = np.random.uniform(-random_weight_threshold, random_weight_threshold, 3)
+
+for epoch in range(epochs):
+    # Go through all the data.
+    for t in zip(train_data, train_data_y):
+        # Predict the y-value based on the weights.
+        predicted_y = threshold_activation_function(weigths.dot(t[0]))
+        # The size of the input data is the size of the weights that must be updated.
+        for entry in range(3):
+            weigths[entry] = weigths[entry] - (
+                learning_rate * (predicted_y - t[1]) * t[0][entry]
+            )
+
+# Test on training data if the output is correct.
+# In this case, there is no data to test.
+for t in zip(train_data, train_data_y):
+    predicted_y = threshold_activation_function(weigths.dot(t[0]))
+    print(t)
+    print(predicted_y)
-- 
GitLab