#!/usr/bin/env python
# coding: utf-8
# # Catastrophic Cancellation
# In[5]:
import numpy as np
# Let's make two numbers with very similar magnitude:
# In[6]:
x = 1.48234
y = 1.48235
# Now let's compute their difference in double precision:
# In[7]:
x_dbl = np.float64(x)
y_dbl = np.float64(y)
diff_dbl = x_dbl-y_dbl
print(repr(diff_dbl))
# * What would the correct result be?
# * What has happened here?
# -------------
# Can you predict what will happen in single precision?
# In[8]:
x_sng = np.float32(x)
y_sng = np.float32(y)
diff_sng = x_sng-y_sng
print(diff_sng)
# In[ ]: