The idea is that the text file has 150 rows where each row is a string of 1024 bits (a representation of a 32x32 image).
What i want to achieve is to have an array of 150 elements where every element is an array of size 1024.
By trying the code below i get an array of 150 elements with inf value. Is there a way to convert those values to vectors using numpy's loadtxt directly.
Thank you in advance!
import numpy as np
data = np.loadtxt("digits.txt")
If each line is exactly the same length and contains only the characters 0 and 1, you can use numpy.genfromtxt, with delimiter=1. When the argument delimiter is a single integer, genfromtxt treats each line as a sequence of fixed-width fields. The value given to delimiter specifies the field width.
For example, suppose the file 01.txt contains
0001
1010
1111
0000
1001
Here's how you can use genfromtxt to read that into a NumPy integer array with shape (5, 4):
In [2]: import numpy as np
In [3]: data = np.genfromtxt('01.txt', delimiter=1, dtype=np.int8)
In [4]: data
Out[4]:
array([[0, 0, 0, 1],
[1, 0, 1, 0],
[1, 1, 1, 1],
[0, 0, 0, 0],
[1, 0, 0, 1]], dtype=int8)
supposed your text file contains 128 characters in each line (excluding newline character), each character representing 1 byte / 8 bits, you could use
data = np.loadtxt(file, dtype=np.str)
bits_arr = []
for line in data:
byte_arr = np.frombuffer(line.encode('UTF-8'), dtype=np.uint8) # UTF-8 assumed
bits_arr.append(np.unpackbits(byte_arr).reshape(32,32))
bits_arr will then contain 1 "32x32 bitmap" for each line. Note that reshape(32,32) will fail if an invalid number of bytes (!=128) is read in a line.
Sidenote: it is probably more efficient here to use a simple readlines() instead of carrying around all the overhead of np.loadtxt since you actually don't use what this function can do for you. The code could therefore be simplified to
bits_arr = []
with open(file, 'rb') as binfile:
line = binfile.readline().strip() # strip to remove newline char
byte_arr = np.frombuffer(line, dtype=np.uint8)
bits_arr.append(np.unpackbits(byte_arr).reshape(32,32))
0and1? If not, how are the binary vectors represented in the file?.read(128)for each 1024 bits... no newlines ("rows") needed.loadtxtdocs? Default dtype is float. If you have a 1000 digits without delimiter, it tries to make one number from that. Theinfvalue is likely.