linear regression

linear_regression.py

@@ -0,0 +1,60 @@
+import pandas as pd
+import matplotlib.pyplot as plt
+
+def predict(x, w, b):
+  return x * w + b
+
+def loss(x, y):
+  return (y - x) ** 2
+
+def mse(predicted, actual):
+  sum = 0.0
+  for i, x in enumerate(predicted):
+    sum += loss(x, actual[i])
+  return sum / len(predicted)
+
+def slope_weight(input, predicted, actual):
+  sum = 0.0
+  for i, x in enumerate(predicted):
+    sum += (actual[i] - x) * input[i]
+  return sum / len(predicted) * 2
+
+def slope_bias(predicted, actual):
+  sum = 0.0
+  for i, x in enumerate(predicted):
+    sum += (actual[i] - x)
+  return sum / len(predicted) * 2
+
+data  = pd.read_csv("california_housing_train.csv")
+
+w = 0
+b = 0
+prev = 0
+delta = 10
+
+beta = 0.01
+
+median_income = data['median_income']
+median_house_value = data['median_house_value']
+
+mu_input = median_income.mean()
+sigma_input = median_income.std()
+
+mu_actual = median_house_value.mean()
+sigma_actual = median_house_value.std()
+
+median_income = (median_income - mu_input) / sigma_input
+median_house_value = (median_house_value - mu_actual) / sigma_actual
+
+while delta > 1e-6:
+  predicted = []
+  actual = median_house_value.tolist()
+  for sample in median_income.tolist():
+    predicted.append(predict(sample, w, b))
+  delta = abs(mse(predicted, actual) - prev)
+  
+  w += slope_weight(median_income.tolist(), predicted, actual) * beta
+  b += slope_bias(predicted, actual) * beta
+  prev = mse(predicted, actual)
+
+# `w` and `b` now are our weight and bias