« first day (3924 days earlier)      last day (78 days later) » 

2:39 AM
@Rob hey…
I raised a couple of flags on a user and it seems the user admitted in a post that they closed…
but I will presume you can track down that flag nevertheless.
 
 
2 hours later…
4:29 AM
@satan29 It is an example of a tensor eating two vectors to produce a scalar.
 
@JohnRennie I see
 
@satan29 Second rank tensors do that! :-)
 
 
8 hours later…
12:31 PM
hi
WHat happened to all the stars!?!?!?!
 
@satan29 are they invisible?
 
yes
 
I've been getting this problem too sometimes, they usually load after some time
have a lot of your messages been timing out as well?
 
YES OH GOD
for 2 days now. Its a menace
 
yup
 
12:36 PM
the stars are blinking out, one by one
must be the universe ending
 
12:49 PM
any idea how to get the array in a more...readable format?
 
@satan29 redefine __str__ or __repr__ of the object you're printing there
 
I dont follow...
Something like this can be done by pd.read_csv() and then printing it but we dont have csv file here...
 
the behaviour of a Python object when passed to print is defined by its __str__ or __repr__ methods
 
oh
 
what you see there is the default behaviour - the object is just printing its list of attributes as { <attribute_name> : print(attribute), ... }
so it has some attribute called data, and then that is an array that prints itself like that messy array
 
12:55 PM
yeah
 
you can try passing it to pprint and see whether that improves things
but otherwise you have little choice but to implement the pretty-printing yourself by looping over data and printing it how you like
 
 
1 hour later…
2:16 PM
pandas and pd.DataFrame(boston) may get you there, though you might need mappings for the data columns for it to actually look nice
 
 
3 hours later…
5:23 PM
How can I normalise each column of a numpy array?
by normalise, i mean, $xi-->xi-\mu / \sigma$
where $\mu$ and $\sigma$ are mean and SD of that column
 
Well `array.std()` and `array.mean()` are certainly available to do it
If you're using sklearn, it also has some preprocessing functions that you can use on their own or in a pipeline
 
@DanielUnderwood but that will give a single value for the entire array, no?
I want to do it column by column
 
Ah I didn't catch that part. I'm not sure off the top of my head, but I think it would be the axis parameter to those two functions
 
Another issue is that whenever i try to print a certain column of the array, the values are in scientific format (eg 1.024 e 02) which is making things cumbersome
@DanielUnderwood My idea is to use what you suggested after extracting the particular columns in a loop
 
5:41 PM
Yeah you could do it that way by slicing in a loop as well. A python loop will be a bit slower than having numpy do the whole thing at once, but that likely won't matter if the array is small
I think the scientific notation thing might be an option somewhere
 
@DanielUnderwood hmm right, but right now I just want to get the whole thing done first, Ill look to optimize later
 
Compute the mean and stddev of the matrix using the respective numpy functions with axis=1
Then do x = x - mean/stddev
Broadcasting should take care of the rest
 
okay hang on, I fixed the scientific notation problem, however for some reason I dont think the numpy array isnt showing the correct values
import sklearn
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.datasets import load_boston



boston=load_boston()
#print(boston)
X , y = load_boston(return_X_y = True)
y = y.reshape(506,1)

df_x=pd.DataFrame(boston.data,columns=boston.feature_names)
df_y=pd.DataFrame(boston.target)

print(df_x)

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)

np.set_printoptions(precision=3,suppress=True)
printing the dataframe df_x and the numpy array X-train gives different values...
 
Does train_test_split shuffle by default?
 
5:57 PM
ah damn
the dafult for shuffle is true hmm
@NiharKarve what exactly do you mean by broadcasting?
and with axis =1, stdev and mean would be arrays themselves, so how can you divide by an array? Or does "broadcasting" interpret this in the way we are looking for?
 
Yes, numpy should divide them element-wise
 
Ah I see
 
This yields a length M vector, which you can't directly add to an MxN matrix
But broadcasting comes in again - it adds this same row vector to each row in the matrix, which I think is what you are looking for
 
wait, I didn't catch that last part
 
Try this: jakevdp.github.io/PythonDataScienceHandbook/… (sorry I'm on phone right now so I can't really type fast or verify the code)
 
6:10 PM
@NiharKarve this should be axis 0 ?
 
Possibly, I can never remember whether rows or columns come first
 
6:27 PM
my last question for today
suppose you have an array
1 2 3
4 5 6
7 8 9
and you need to add a bias vector b at the start. (initialised to all 1s initially). So basically:
1 1 2 3
1 4 5 6
1 7 8 9
How exactly can we do this? It has something to do with either np.hstack() or vstack but I can't understand either of them properly..
 
It may be one of np.stack or np.concatenate?
I usually just keep trying until I manage to get it right
 
 
1 hour later…
fqq
7:40 PM
np.hstack([np.ones((3,1)),a])

where a is the original 3x3 array
 

« first day (3924 days earlier)      last day (78 days later) »