Parallelize a list item append to dict using multiprocessing - python-3.x

I have a large list containing strings. I wish to create a dict from this list such that:
list = [str1, str2, str3, ....]
dict = {str1:len(str1), str2:len(str2), str3:len(str3),.....}
My go to solution was a for loop but its taking too much time (my list contains almost 1M elements):
for i in list:
d[i] = len(i)
I wish to use the multiprocessing module in python in order to leverage all cores and reduce the time taken for the process to execute. I have come across some crude examples involving manager module to share dict between different processes but am unable to implement it. Any help would be appreciated!

I don't know if using multiple process will be faster, but it's an interesting experiment.
General flow:
Create list of random words
Split list into segments, one segment per process
Run processes, pass segment as parameter
Merge result dictionaries to single dictionary
Try this code:
import concurrent.futures
import random
from multiprocessing import Process, freeze_support
def todict(lst):
print(f'Processing {len(lst)} words')
return {e:len(e) for e in lst} # convert list to dictionary
if __name__ == '__main__':
freeze_support() # needed for Windows
# create random word list - max 15 chars
letters = [chr(x) for x in range(65,65+26)] # A-Z
words = [''.join(random.sample(letters,random.randint(1,15))) for w in range(10000)] # 10000 words
words = list(set(words)) # remove dups, count will drop
print(len(words))
########################
cpucnt = 4 # process count to use
# split word list for each process
wl = len(words)//cpucnt + 1 # word count per process
lstsplit = []
for c in range(cpucnt):
lstsplit.append(words[c*wl:(c+1)*wl]) # create word list for each process
# start processes
with concurrent.futures.ProcessPoolExecutor(max_workers=cpucnt) as executor:
procs = [executor.submit(todict, lst) for lst in lstsplit]
results = [p.result() for p in procs] # block until results are gathered
# merge results to single dictionary
dd = {}
for r in results:
dd.update(r)
print(len(dd)) # confirm match word count
with open('dd.txt','w') as f: f.write(str(dd)) # write dictionary to text file

Related

Is there any ways to make this more efficient?

I have 24 more attempts to submit this task. I spent hours and my brain does not work anymore. I am a beginner with Python can you please help to figure out what is wrong? I would love to see the correct code if possible.
Here is the task itself and the code I wrote below.
Note that you can have access to all standard modules/packages/libraries of your language. But there is no access to additional libraries (numpy in python, boost in c++, etc).
You are given a content of CSV-file with information about set of trades. It contains the following columns:
TIME - Timestamp of a trade in format Hour:Minute:Second.Millisecond
PRICE - Price of one share
SIZE - Count of shares executed in this trade
EXCHANGE - The exchange that executed this trade
For each exchange find the one minute-window during which the largest number of trades took place on this exchange.
Note that:
You need to send source code of your program.
You have only 25 attempts to submit a solutions for this task.
You have access to all standart modules/packages/libraries of your language. But there is no access to additional libraries (numpy in python, boost in c++, etc).
Input format
Input contains several lines. You can read it from standart input or file “trades.csv”
Each line contains information about one trade: TIME, PRICE, SIZE and EXCHANGE. Numbers are separated by comma.
Lines are listed in ascending order of timestamps. Several lines can contain the same timestamp.
Size of input file does not exceed 5 MB.
See the example below to understand the exact input format.
Output format
If input contains information about k exchanges, print k lines to standart output.
Each line should contain the only number — maximum number of trades during one minute-window.
You should print answers for exchanges in lexicographical order of their names.
Sample
Input Output
09:30:01.034,36.99,100,V
09:30:55.000,37.08,205,V
09:30:55.554,36.90,54,V
09:30:55.556,36.91,99,D
09:31:01.033,36.94,100,D
09:31:01.034,36.95,900,V
2
3
Notes
In the example four trades were executed on exchange “V” and two trades were executed on exchange “D”. Not all of the “V”-trades fit in one minute-window, so the answer for “V” is three.
X = []
with open('trades.csv', 'r') as tr:
for line in tr:
line = line.strip('\xef\xbb\xbf\r\n ')
X.append(line.split(','))
dex = {}
for item in X:
dex[item[3]] = []
for item in X:
dex[item[3]].append(float(item[0][:2])*60.+float(item[0][3:5])+float(item[0][6:8])/60.+float(item[0][9:])/60000.)
for item in dex:
count = 1
ccount = 1
if dex[item][len(dex[item])-1]-dex[item][0] <1:
count = len(dex[item])
else:
for t in range(len(dex[item])-1):
for tt in range(len(dex[item])-t-1):
if dex[item][tt+t+1]-dex[item][t] <1:
ccount += 1
else: break
if ccount>count:
count=ccount
ccount=1
print(count)
First of all it is not necessary to use datetime and csv modules for such a simple case (like in Ed-Ward's example).
If we remove colon and dot signs from the time strings it could be converted to int() directly - easier way than you tried in your example.
CSV features like dialect and special formatting not used so i suggest to use simple split(",")
Now about efficiency. Efficiency means time complexity.
The more times you go through your array with dates from the beginning to the end, the more complicated the algorithm becomes.
So our goal is to minimize cycles count, best to make only one pass by all rows and especially avoid nested loops and passing through collections from beginning to the end.
For such a task it is better to use deque, instead of tuple or list, because you can pop() first element and append last element with complexity of O(1).
Just append every time for needed exchange to the end of the exchange's queue until difference between current and first elements becomes more than 1 minute. Then just remove first element with popleft() and continue comparison. After whole file done - length of each queue will be the max 1min window.
Example with linear time complexity O(n):
from collections import deque
ex_list = {}
s = open("trades.csv").read().replace(":", "").replace(".", "")
for line in s.splitlines():
s = line.split(",")
curr_tm = int(s[0])
curr_ex = s[3]
if curr_ex not in ex_list:
ex_list[curr_ex] = deque()
ex_list[curr_ex].append(curr_tm)
if curr_tm >= ex_list[curr_ex][0] + 100000:
ex_list[curr_ex].popleft()
print("\n".join([str(len(ex_list[k])) for k in sorted(ex_list.keys())]))
This code should work:
import csv
import datetime
diff = datetime.timedelta(minutes=1)
def date_calc(start, dates):
for i, date in enumerate(dates):
if date >= start + diff:
return i
return i + 1
exchanges = {}
with open("trades.csv") as csvfile:
reader = csv.reader(csvfile)
for row in reader:
this_exchange = row[3]
if this_exchange not in exchanges:
exchanges[this_exchange] = []
time = datetime.datetime.strptime(row[0], "%H:%M:%S.%f")
exchanges[this_exchange].append(time)
ex_max = {}
for name, dates in exchanges.items():
ex_max[name] = 0
for i, d in enumerate(dates):
x = date_calc(d, dates[i:])
if x > ex_max[name]:
ex_max[name] = x
print('\n'.join([str(ex_max[k]) for k in sorted(ex_max.keys())]))
Output:
2
3
( obviously please check it for yourself before uploading it :) )
I think the issue with your current code is that you don't put the output in lexicographical order of their names...
If you want to use your current code, then here is a (hopefully) fixed version:
X = []
with open('trades.csv', 'r') as tr:
for line in tr:
line = line.strip('\xef\xbb\xbf\r\n ')
X.append(line.split(','))
dex = {}
counts = []
for item in X:
dex[item[3]] = []
for item in X:
dex[item[3]].append(float(item[0][:2])*60.+float(item[0][3:5])+float(item[0][6:8])/60.+float(item[0][9:])/60000.)
for item in dex:
count = 1
ccount = 1
if dex[item][len(dex[item])-1]-dex[item][0] <1:
count = len(dex[item])
else:
for t in range(len(dex[item])-1):
for tt in range(len(dex[item])-t-1):
if dex[item][tt+t+1]-dex[item][t] <1:
ccount += 1
else: break
if ccount>count:
count=ccount
ccount=1
counts.append((item, count))
counts.sort(key=lambda x: x[0])
print('\n'.join([str(x[1]) for x in counts]))
Output:
2
3
I do think you can make your life easier in the future by using Python's standard library, though :)

Convert everything in a dictionary to lower case, then filter on it?

import pandas as pd
import nltk
import os
directory = os.listdir(r"C:\...")
x = []
num = 0
for i in directory:
x.append(pd.read_fwf("C:\\..." + i))
x[num] = x[num].to_string()
So, once I have a dictionary x = [ ] populated by the read_fwf for each file in my directory:
I want to know how to make it so every single character is lowercase. I am having trouble understanding the syntax and how it is applied to a dictionary.
I want to define a filter that I can use to count for a list of words in this newly defined dictionary, e.g.,
list = [bus, car, train, aeroplane, tram, ...]
Edit: Quick unrelated question:
Is pd_read_fwf the best way to read .txt files? If not, what else could I use?
Any help is very much appreciated. Thanks
Edit 2: Sample data and output that I want:
Sample:
The Horncastle boar's head is an early seventh-century Anglo-Saxon
ornament depicting a boar that probably was once part of the crest of
a helmet. It was discovered in 2002 by a metal detectorist searching
in the town of Horncastle, Lincolnshire. It was reported as found
treasure and acquired for £15,000 by the City and County Museum, where
it is on permanent display.
Required output - changes everything in uppercase to lowercase:
the horncastle boar's head is an early seventh-century anglo-saxon
ornament depicting a boar that probably was once part of the crest of
a helmet. it was discovered in 2002 by a metal detectorist searching
in the town of horncastle, lincolnshire. it was reported as found
treasure and acquired for £15,000 by the city and county museum, where
it is on permanent display.
You shouldn't need to use pandas or dictionaries at all. Just use Python's built-in open() function:
# Open a file in read mode with a context manager
with open(r'C:\path\to\you\file.txt', 'r') as file:
# Read the file into a string
text = file.read()
# Use the string's lower() method to make everything lowercase
text = text.lower()
print(text)
# Split text by whitespace into list of words
word_list = text.split()
# Get the number of elements in the list (the word count)
word_count = len(word_list)
print(word_count)
If you want, you can do it in the reverse order:
# Open a file in read mode with a context manager
with open(r'C:\path\to\you\file.txt', 'r') as file:
# Read the file into a string
text = file.read()
# Split text by whitespace into list of words
word_list = text.split()
# Use list comprehension to create a new list with the lower() method applied to each word.
lowercase_word_list = [word.lower() for word in word_list]
print(word_list)
Using a context manager for this is good since it automatically closes the file for you as soon as it goes out of scope (de-tabbed from with statement block). Otherwise you would have to use file.open() and file.read().
I think there are some other benefits to using context managers, but someone please correct me if I'm wrong.
I think what you are looking for is dictionary comprehension:
# Python 3
new_dict = {key: val.lower() for key, val in old_dict.items()}
# Python 2
new_dict = {key: val.lower() for key, val in old_dict.iteritems()}
items()/iteritems() gives you a list of tuples of the (keys, values) represented in the dictionary (e.g. [('somekey', 'SomeValue'), ('somekey2', 'SomeValue2')])
The comprehension iterates over each of these pairs, creating a new dictionary in the process. In the key: val.lower() section, you can do whatever manipulation you want to create the new dictionary.

Split list into randomised ordered sub lists

I would like to improve the below code to split a list of values into two sub lists, which have been randomised and sorted. The below code works, but I'm sure there is a better/cleaner way to do it.
import random
data = list(range(1, 61))
random.shuffle(data)
Intervention = data[:30]
Control = data[30:]
Intervention.sort()
Control.sort()
f = open('Randomised_Groups.txt', 'w')
f.write('Intervention Group = ' + str(Intervention) + '\n' + 'Control Group = ' + str(Control))
f.close()
The expected output is:
Intervention = [1,3,7,9]
Control = [2,4,5,6,8,10]
I think your code is short and clean already. Some changes you can make:
Call sorted() when you slice it.
Intervention = sorted(data[:30])
You can also define both Intervention and Control on one line:
Intervention, Control = data[:30], data[30:]
I would replace the 30 with a variable:
half = len(data)//2
It is safer to open a file with with. That closes the file automatically when indentation stops.
with open('Randomised_Groups.txt', 'w') as f:
...
With the use of f-strings you can make the write statement shorter:
f.write(f'Intervention Group = {Intervention} \nControl Group = {Control}')
All combined:
import random
data = list(range(1, 61))
random.shuffle(data)
half = len(data)//2
Intervention, Control = sorted(data[:half]), sorted(data[half:])
with open('Randomised_Groups.txt', 'w') as f:
f.write(f'Intervention Group = {Intervention}\nControl Group = {Control}')
Something like this might be what you want:
import random
my_rng = [random.randint(0,1) for i in range(60)]
Control = [i for i in range(60) if my_rng[i] == 0]
Intervention = [i for i in range(60) if my_rng[i] == 1]
print(Control)
The idea is to create 60 random 1s or 0s to use as indicators for which list to put each number in. This will only work if you do not need the two lists to be the same length. To get the same length would require changing how my_rng is created in this example.
I have tinkered a bit further and got the lists of the same length:
import random
my_rng = [0 for i in range(30)]
my_rng.extend([1 for i in range(30)])
random.shuffle(my_rng)
Control = [i for i in range(60) if my_rng[i] == 0]
Intervention = [i for i in range(60) if my_rng[i] == 1]
Here, instead of adding randomly 1 or 0 to my_rng I get a list of 30 0s and 30 1s to shuffle, then continue like before.
Here is another solution that is more dynamic using built in random functionality that only creates the lists needed (no extra memory) and would work with lists that contain any type of object (provided that object can be sorted):
import random
def convert_to_random_list(data, num_list):
"""
Takes in the data as one large list and converts it into
[num_list] random sorted lists.
"""
result_lists = [list() for _ in range(num_list)] # two lists
for x in data:
# Using randint we pick which list to insert into
result_lists[random.randint(0, num_list - 1)].append(x)
# You could use list comprehension here with `sorted(...)` but it would take a little extra memory.
for _list in result_lists:
_list.sort()
return result_lists
Can be tested with:
data = list(range(1, 61))
random.shuffle(data)
temp = convert_to_random_list(data, 3)
print(temp)

How to speed up for loop execution using multiprocessing in python

I have two lists. List A contains 500 words. List B contains 10000 words. I am trying to find similar words for List A with respect to B.I am using Spacy's similarity function.
The problem I am facing is that it takes ages to compute. I am new to multiprocessing usage, hence request help.
How do I speed up the execution of the for loop part through multiprocessing in python?
The following is my code.
ListA =['Dell', 'GPU',......] #500 words lists
ListB = ['Docker','Ec2'.......] #10000 words lists
s_words = []
for token1 in ListB:
list_to_sort = []
for token2 in ListA:
list_to_sort.append((token1, token2,nlp(str(token1)).similarity(nlp(str(token2)))))
sorted_list = sorted(list_to_sort, key = itemgetter(2), reverse=True)[0][:2]
s_words.append(sorted_list)
You can use multiprocessing package. This I hope will reduce your time significantly. See here for a sample code.
Have you tried nlp.pipe()?
You could do something like this:
from operator import itemgetter
import spacy
nlp = spacy.load("en_core_web_lg")
ListA = ['Apples', 'Monkey'] # 500 words lists
ListB = ['Grapefruit', 'Ape', 'Oranges', 'Banana'] # 10000 words lists
s_words = []
docs_a = nlp.pipe(ListA)
docs_b = list(nlp.pipe(ListB))
for token1 in docs_a:
list_to_sort = []
for token2 in docs_b:
list_to_sort.append((token1.text, token2.text, token1.similarity(token2)))
sorted_list = sorted(list_to_sort, key=itemgetter(2), reverse=True)[0][:2]
s_words.append(sorted_list)
print(s_words)
That should already speed things up for you. The function nlp.pipe() also has the parameter n_process which might be what you're looking for.

Analysis of Eye-Tracking data in python (Eye-link)

I have data from eye-tracking (.edf file - from Eyelink by SR-research). I want to analyse it and get various measures such as fixation, saccade, duration, etc.
Is there an existing package to analyse Eye-Tracking data?
Thanks!
At least for importing the .edf-file into a pandas DF, you can use the following package by Niklas Wilming: https://github.com/nwilming/pyedfread/tree/master/pyedfread
This should already take care of saccades and fixations - have a look at the readme. Once they're in the data frame, you can apply whatever analysis you want to it.
pyeparse seems to be another (yet currently unmaintained as it seems) library that can be used for eyelink data analysis.
Here is a short excerpt from their example:
import numpy as np
import matplotlib.pyplot as plt
import pyeparse as pp
fname = '../pyeparse/tests/data/test_raw.edf'
raw = pp.read_raw(fname)
# visualize initial calibration
raw.plot_calibration(title='5-Point Calibration')
# create heatmap
raw.plot_heatmap(start=3., stop=60.)
EDIT: After I posted my answer I found a nice list compiling lots of potential tools for eyelink edf data analysis: https://github.com/davebraze/FDBeye/wiki/Researcher-Contributed-Eye-Tracking-Tools
Hey the question seems rather old but maybe I can reactivate it, because I am currently facing the same situation.
To start I recommend to convert your .edf to an .asc file. In this way it is easier to read it to get a first impression.
For this there exist many tools, but I used the SR-Research Eyelink Developers Kit (here).
I don't know your setup but the Eyelink 1000 itself detects saccades and fixation. I my case in the .asc file it looks like that:
SFIX L 10350642
10350642 864.3 542.7 2317.0
...
...
10350962 863.2 540.4 2354.0
EFIX L 10350642 10350962 322 863.1 541.2 2339
SSACC L 10350964
10350964 863.4 539.8 2359.0
...
...
10351004 683.4 511.2 2363.0
ESACC L 10350964 10351004 42 863.4 539.8 683.4 511.2 5.79 221
The first number corresponds to the timestamp, the second and third to x-y coordinates and the last is your pupil diameter (what the last numbers after ESACC are, I don't know).
SFIX -> start fixation
EFIX -> end fixation
SSACC -> start saccade
ESACC -> end saccade
You can also check out PyGaze, I haven't worked with it, but searching for a toolbox, this one always popped up.
EDIT
I found this toolbox here. It looks cool and works fine with the example data, but sadly does not work with mine
EDIT No 2
Revisiting this question after working on my own Eyetracking data I thought I might share a function wrote, to work with my data:
def eyedata2pandasframe(directory):
'''
This function takes a directory from which it tries to read in ASCII files containing eyetracking data
It returns eye_data: A pandas dataframe containing data from fixations AND saccades fix_data: A pandas dataframe containing only data from fixations
sac_data: pandas dataframe containing only data from saccades
fixation: numpy array containing information about fixation onsets and offsets
saccades: numpy array containing information about saccade onsets and offsets
blinks: numpy array containing information about blink onsets and offsets
trials: numpy array containing information about trial onsets
'''
eye_data= []
fix_data = []
sac_data = []
data_header = {0: 'TimeStamp',1: 'X_Coord',2: 'Y_Coord',3: 'Diameter'}
event_header = {0: 'Start', 1: 'End'}
start_reading = False
in_blink = False
in_saccade = False
fix_timestamps = []
sac_timestamps = []
blink_timestamps = []
trials = []
sample_rate_info = []
sample_rate = 0
# read the file and store, depending on the messages the data
# we have the following structure:
# a header -- every line starts with a '**'
# a bunch of messages containing information about callibration/validation and so on all starting with 'MSG'
# followed by:
# START 10350638 LEFT SAMPLES EVENTS
# PRESCALER 1
# VPRESCALER 1
# PUPIL AREA
# EVENTS GAZE LEFT RATE 500.00 TRACKING CR FILTER 2
# SAMPLES GAZE LEFT RATE 500.00 TRACKING CR FILTER 2
# followed by the actual data:
# normal data --> [TIMESTAMP]\t [X-Coords]\t [Y-Coords]\t [Diameter]
# Start of EVENTS [BLINKS FIXATION SACCADES] --> S[EVENTNAME] [EYE] [TIMESTAMP]
# End of EVENTS --> E[EVENT] [EYE] [TIMESTAMP_START]\t [TIMESTAMP_END]\t [TIME OF EVENT]\t [X-Coords start]\t [Y-Coords start]\t [X_Coords end]\t [Y-Coords end]\t [?]\t [?]
# Trial messages --> MSG timestamp\t TRIAL [TRIALNUMBER]
try:
with open(directory) as f:
csv_reader = csv.reader(f, delimiter ='\t')
for i, row in enumerate (csv_reader):
if any ('RATE' in item for item in row):
sample_rate_info = row
if any('SYNCTIME' in item for item in row): # only start reading after this message
start_reading = True
elif any('SFIX' in item for item in row): pass
#fix_timestamps[0].append (row)
elif any('EFIX' in item for item in row):
fix_timestamps.append ([row[0].split(' ')[4],row[1]])
#fix_timestamps[1].append (row)
elif any('SSACC' in item for item in row):
#sac_timestamps[0].append (row)
in_saccade = True
elif any('ESACC' in item for item in row):
sac_timestamps.append ([row[0].split(' ')[3],row[1]])
in_saccade = False
elif any('SBLINK' in item for item in row): # stop reading here because the blinks contain NaN
# blink_timestamps[0].append (row)
in_blink = True
elif any('EBLINK' in item for item in row): # start reading again. the blink ended
blink_timestamps.append ([row[0].split(' ')[2],row[1]])
in_blink = False
elif any('TRIAL' in item for item in row):
# the first element is 'MSG', we don't need it, then we split the second element to seperate the timestamp and only keep it as an integer
trials.append (int(row[1].split(' ')[0]))
elif start_reading and not in_blink:
eye_data.append(row)
if in_saccade:
sac_data.append(row)
else:
fix_data.append(row)
# drop the last data point, because it is the 'END' message
eye_data.pop(-1)
sac_data.pop(-1)
fix_data.pop(-1)
# convert every item in list into a float, substract the start of the first trial to set the start of the first video to t0=0
# then devide by 1000 to convert from milliseconds to seconds
for row in eye_data:
for i, item in enumerate (row):
row[i] = float (item)
for row in fix_data:
for i, item in enumerate (row):
row[i] = float (item)
for row in sac_data:
for i, item in enumerate (row):
row[i] = float (item)
for row in fix_timestamps:
for i, item in enumerate (row):
row [i] = (float(item)-trials[0])/1000
for row in sac_timestamps:
for i, item in enumerate (row):
row [i] = (float(item)-trials[0])/1000
for row in blink_timestamps:
for i, item in enumerate (row):
row [i] = (float(item)-trials[0])/1000
sample_rate = float (sample_rate_info[4])
# convert into pandas fix_data Frames for a better overview
eye_data = pd.DataFrame(eye_data)
fix_data = pd.DataFrame(fix_data)
sac_data = pd.DataFrame(sac_data)
fix_timestamps = pd.DataFrame(fix_timestamps)
sac_timestamps = pd.DataFrame(sac_timestamps)
trials = np.array(trials)
blink_timestamps = pd.DataFrame(blink_timestamps)
# rename header for an even better overview
eye_data = eye_data.rename(columns=data_header)
fix_data = fix_data.rename(columns=data_header)
sac_data = sac_data.rename(columns=data_header)
fix_timestamps = fix_timestamps.rename(columns=event_header)
sac_timestamps = sac_timestamps.rename(columns=event_header)
blink_timestamps = blink_timestamps.rename(columns=event_header)
# substract the first timestamp of trials to set the start of the first video to t0=0
eye_data.TimeStamp -= trials[0]
fix_data.TimeStamp -= trials[0]
sac_data.TimeStamp -= trials[0]
trials -= trials[0]
trials = trials /1000 # does not work with trials/=1000
# devide TimeStamp to get time in seconds
eye_data.TimeStamp /=1000
fix_data.TimeStamp /=1000
sac_data.TimeStamp /=1000
return eye_data, fix_data, sac_data, fix_timestamps, sac_timestamps, blink_timestamps, trials, sample_rate
except:
print ('Could not read ' + str(directory) + ' properly!!! Returned empty data')
return eye_data, fix_data, sac_data, fix_timestamps, sac_timestamps, blink_timestamps, trials, sample_rate
Hope it helps you guys. Some parts of the code you may need to change, like the index where to split the strings to get the crutial information about event on/offsets. Or you don't want to convert your timestamps into seconds or do not want to set the onset of your first trial to 0. That is up to you.
Additionally in my data we sent a message to know when we started measuring ('SYNCTIME') and I had only ONE condition in my experiment, so there is only one 'TRIAL' message
Cheers

Resources