Greetings,
I have a 2D numpy array
ARR
storing a bunch of floats. I want to print it row by row to a .csv file, but rounded to 3 significant digits.
Is there a simple way to do this?
This is what I currently have. It prints the array without rounding.
with open('./results', 'w') as f:
writer = csv.writer(f, delimiter=" ")
for row in ARR:
writer.writerow(row) Cheers!
Take into Pandas the can use
.to_csv with
float_format.
import numpy as np
import pandas as pd
arr_rand = np.random.rand(3,4)
df = pd.DataFrame(arr_rand)
>>> df
0 1 2 3
0 0.440761 0.337663 0.372010 0.892344
1 0.952017 0.747404 0.787696 0.338794
2 0.690075 0.061731 0.541146 0.968825
>>> print(df.to_csv(index=False, header=False, line_terminator='\n', float_format='%.3f').strip())
0.441,0.338,0.372,0.892
0.952,0.747,0.788,0.339
0.690,0.062,0.541,0.969
The simplest way might be to use pandas.
import numpy as np
import pandas as pd
# Make a float numpy array
data = np.array([[a/b for b in range(1,11)] for a in range(1,11)])
# Write to csv file
df = pd.DataFrame(data)
df.to_csv("test.csv", index=False, header=False)Output:
1.0,0.5,0.3333333333333333,0.25,0.2,0.16666666666666666,...
Another way to do this is convert all the numbers to strings and write the strings to the csv file.
Thank you, you two!
I just found another way. Apparently, numpy offers a rounding function for their arrays, so one can just call
ARR.round(decim)
where
ARR
is a numpy array to be rounded to
decim
signifiant figures.
Then my snippet from above can be used.
I was solving the wrong problem. I saw CSV and rounding and immediately jumped to the conclusion that you were not happy about losing precision when saving your numpy array to a file. That is the complaint in 99.9% of posts that include "csv" and "rounding" in their topic. Sorry for wasting your time.
Curious though, why do you want to round floats to 3 digits in a csv file? It's not like csv is a presentation format.
(Nov-14-2022, 08:25 PM)deanhystad Wrote: [ -> ]Curious though, why do you want to round floats to 3 digits in a csv file? It's not like csv is a presentation format.
I want to lose precision so that the file size does not get unnecessarily large. Saving, say, 5 digits instead of the usual 15 saves a factor 3 then. I work in scientific computing and result files often store millions of floats, so stuff like this pays off.
Why not use a binary format? Numpy loves binary. A float is 4 bytes. More precision than 6 ascii characters (need a separator). Much faster to read and write.
Or is this some kind of intermediary that has to be readable by programs where numpy is not an option?