Question

I’m using numpy loadtxt function to read in a large set of data. The data appears to be rounded off. for example: The number in the text file is -3.79000000000005E+01 but numpy reads the number in as -37.9. I’ve set the dypte to np.float64 in the loadtxt call. Is there anyway to keep the precision of the original data file?

Answer 1

loadtxt is not rounding the number. What you are seeing is the way NumPy chooses to print the array:

In [80]: import numpy as np

In [81]: x = np.loadtxt('test.dat', dtype = np.float64)

In [82]: print(x)
-37.9

The actual value is the np.float64 closest to the value inputted.

In [83]: x
Out[83]: array(-37.9000000000005)

---

Or, in the more likely instance that you have a higher dimensional array,

In [2]: x = np.loadtxt('test.dat', dtype = np.float64)

If the repr of x looks truncated:

In [3]: x
Out[3]: array([-37.9, -37.9])

you can use np.set_printoptions to get higher precision:

In [4]: np.get_printoptions()
Out[4]: 
{'edgeitems': 3,
 'infstr': 'inf',
 'linewidth': 75,
 'nanstr': 'nan',
 'precision': 8,
 'suppress': False,
 'threshold': 1000}

In [5]: np.set_printoptions(precision = 17)

In [6]: x
Out[6]: array([-37.90000000000050306, -37.90000000000050306])

(Thanks to @mgilson for pointing this out.)