I'm trying to apply a preprocessor to remove html tags from the fileds of a Tensorflow dataset. I use this in the implementation of t5 which uses segio.
in my_preprocessor if I use text = row[field_index], it works okay and in training, it prints several rows and continues, but if I use text = normalize_text(row[field_index]) it hangs on an infinite or long loop as if it wants to apply it on the whole dataset rows!
def normalize_text(text):
"""Lowercase and remove quotes and tags from a TensorFlow string."""
text = tf.strings.lower(text)
text = tf.strings.regex_replace(text,"'(.*)'", r"\1")
text = tf.strings.regex_replace(text,"<[^>]+>", " ")
return text
def make_add_field_names_preprocessor(
field_names: Sequence[str], field_indices: Optional[Sequence[int]] = None,
) -> Callable:
def my_preprocessor(ds):
def to_inputs_and_targets(*row):
ret = {}
count=0
tf.print("=======================")
for field_name, field_index in zip(field_names, field_indices):
# if I use this, it works okay
text = row[field_index]
# if I use the following it falls to a long or endless loop
text = normalize_text(row[field_index])
tf.print(count, "Row:",text)
ret[field_name] = text
count+=1
return ret
return ds.map(to_inputs_and_targets,
num_parallel_calls=tf.data.experimental.AUTOTUNE)
return my_preprocessor
Update: The preprocessor gets two field names, "inputs" for row[1], and "targets" for row[2]. I noticed that the problem exists many when I call normalize_text on row[2]
Related
I am trying to scrape data from a word document available at:-
https://dl.dropbox.com/s/pj82qrctzkw9137/HE%20Distributors.docx
I need to scrape the Name, Address, City, State, and Email ID. I am able to scrape the E-mail using the below code.
import docx
content = docx.Document('HE Distributors.docx')
location = []
for i in range(len(content.paragraphs)):
stat = content.paragraphs[i].text
if 'Email' in stat:
location.append(i)
for i in location:
print(content.paragraphs[i].text)
I tried to use the steps mentioned:
How to read data from .docx file in python pandas?
I need to convert this into a data frame with all the columns mentioned above.
Still facing issues with the same.
There are some inconsistencies in the document - phone numbers starting with Tel: sometimes, and Tel.: other times, and even Te: once, and I noticed one of the emails is just in the last line for that distributor without the Email: prefix, and the State isn't always in the last line.... Still, for the most part, most of the data can be extracted with regex and/or splits.
The distributors are separated by empty lines, and the names are in a different color - so I defined this function to get the font color of any paragraph from its xml:
# from bs4 import BeautifulSoup
def getParaColor(para):
try:
return BeautifulSoup(
para.paragraph_format.element.xml, 'xml'
).find('color').get('w:val')
except:
return ''
The try...except hasn't been necessary yet, but just in case...
(The xml is actually also helpful for double-checking that .text hasn't missed anything - in my case, I noticed that the email for Shri Adhya Educational Books wasn't getting extracted.)
Then, you can process the paragraphs from docx.Document with a function like:
# import re
def splitParas(paras):
ptc = [(
p.text, getParaColor(p), p.paragraph_format.element.xml
) for p in paras]
curSectn = 'UNKNOWN'
splitBlox = [{}]
for pt, pc, px in ptc:
# double-check for missing text
xmlText = BeautifulSoup(px, 'xml').text
xmlText = ' '.join([s for s in xmlText.split() if s != ''])
if len(xmlText) > len(pt): pt = xmlText
# initiate
if not pt:
if splitBlox[-1] != {}:
splitBlox.append({})
continue
if pc == '20752E':
curSectn = pt.strip()
continue
if splitBlox[-1] == {}:
splitBlox[-1]['section'] = curSectn
splitBlox[-1]['raw'] = []
splitBlox[-1]['Name'] = []
splitBlox[-1]['address_raw'] = []
# collect
splitBlox[-1]['raw'].append(pt)
if pc == 'D12229':
splitBlox[-1]['Name'].append(pt)
elif re.search("^Te.*:.*", pt):
splitBlox[-1]['tel_raw'] = re.sub("^Te.*:", '', pt).strip()
elif re.search("^Mob.*:.*", pt):
splitBlox[-1]['mobile_raw'] = re.sub("^Mob.*:", '', pt).strip()
elif pt.startswith('Email:') or re.search(".*[#].*[.].*", pt):
splitBlox[-1]['Email'] = pt.replace('Email:', '').strip()
else:
splitBlox[-1]['address_raw'].append(pt)
# some cleanup
if splitBlox[-1] == {}: splitBlox = splitBlox[:-1]
for i in range(len(splitBlox)):
addrsParas = splitBlox[i]['address_raw'] # for later
# join lists into strings
splitBlox[i]['Name'] = ' '.join(splitBlox[i]['Name'])
for k in ['raw', 'address_raw']:
splitBlox[i][k] = '\n'.join(splitBlox[i][k])
# search address for City, State and PostCode
apLast = addrsParas[-1].split(',')[-1]
maybeCity = [ap for ap in addrsParas if '–' in ap]
if '–' not in apLast:
splitBlox[i]['State'] = apLast.strip()
if maybeCity:
maybePIN = maybeCity[-1].split('–')[-1].split(',')[0]
maybeCity = maybeCity[-1].split('–')[0].split(',')[-1]
splitBlox[i]['City'] = maybeCity.strip()
splitBlox[i]['PostCode'] = maybePIN.strip()
# add mobile to tel
if 'mobile_raw' in splitBlox[i]:
if 'tel_raw' not in splitBlox[i]:
splitBlox[i]['tel_raw'] = splitBlox[i]['mobile_raw']
else:
splitBlox[i]['tel_raw'] += (', ' + splitBlox[i]['mobile_raw'])
del splitBlox[i]['mobile_raw']
# split tel [as needed]
if 'tel_raw' in splitBlox[i]:
tel_i = [t.strip() for t in splitBlox[i]['tel_raw'].split(',')]
telNum = []
for t in range(len(tel_i)):
if '/' in tel_i[t]:
tns = [t.strip() for t in tel_i[t].split('/')]
tel1 = tns[0]
telNum.append(tel1)
for tn in tns[1:]:
telNum.append(tel1[:-1*len(tn)]+tn)
else:
telNum.append(tel_i[t])
splitBlox[i]['Tel_1'] = telNum[0]
splitBlox[i]['Tel'] = telNum[0] if len(telNum) == 1 else telNum
return splitBlox
(Since I was getting font color anyway, I decided to add another
column called "section" to put East/West/etc in. And I added "PostCode" too, since it seems to be on the other side of "City"...)
Since "raw" is saved, any other value can be double checked manually at least.
The function combines "Mobile" into "Tel" even though they're extracted with separate regex.
I'd say "Tel_1" is fairly reliable, but some of the inconsistent patterns mean that other numbers in "Tel" might come out incorrect if they were separated with '/'.
Also, "Tel" is either a string or a list of strings depending on how many numbers there were in "tel_raw".
After this, you can just view as DataFrame with:
#import docx
#import pandas
content = docx.Document('HE Distributors.docx')
# pandas.DataFrame(splitParas(content.paragraphs)) # <--all Columns
pandas.DataFrame(splitParas(content.paragraphs))[[
'section', 'Name', 'address_raw', 'City',
'PostCode', 'State', 'Email', 'Tel_1', 'tel_raw'
]]
I'm trying to make the for loop append on all the data set which is from the range of (0,8675) but it keeps getting error just working with 1 row
def convert_emoticons(text):
for emot in EMOTICONS:
text = re.sub(u'('+emot+')', "_".join(EMOTICONS[emot].replace(",","").split()), text)
return text
text = (dataset['posts'][0])
convert_emoticons(text)
Try using just emot:
def convert_emoticons(text):
for emot in EMOTICONS:
text = re.sub(u'('+emot+')', "_".join(emot.replace(",","").split()), text)
return text
text = (dataset['posts'][0])
convert_emoticons(text)
I have a task to check if page is in English, and the following code is not compiling because it writes:
expected string or bytes-like object
train = pd.read_csv('UDSclub/wikipedia_train.csv')
def get_language(page):
res = re.search('[a-z][a-z].wikipedia.org', page)
if res:
return res.group(0)[0:2]
return 'na'
train['language'].apply(get_language)
I am trying to create new column in DataFrame where it will be marked if page is in English of not
Try this:
train = pd.read_csv('UDSclub/wikipedia_train.csv')
def get_language(x):
res = re.search('[a-z][a-z].wikipedia.org', x['page'])
if res:
language = res.group(0)[0:2]
language = 'na'
x['language'] = language
return x
train.apply(get_language)
Im working on a small project of retrieving information about books from the Google Books API using Python 3. For this i make a call to the API, read out the variables and store those in a list. For a search like "linkedin" this works perfectly. However when i enter "Google", it reads the second title from the JSON input. How can this happen?
Please find my code below (Google_Results is the class I use to initialize the variables):
import requests
def Book_Search(search_term):
parms = {"q": search_term, "maxResults": 3}
r = requests.get(url="https://www.googleapis.com/books/v1/volumes", params=parms)
print(r.url)
results = r.json()
i = 0
for result in results["items"]:
try:
isbn13 = str(result["volumeInfo"]["industryIdentifiers"][0]["identifier"])
isbn10 = str(result["volumeInfo"]["industryIdentifiers"][1]["identifier"])
title = str(result["volumeInfo"]["title"])
author = str(result["volumeInfo"]["authors"])[2:-2]
publisher = str(result["volumeInfo"]["publisher"])
published_date = str(result["volumeInfo"]["publishedDate"])
description = str(result["volumeInfo"]["description"])
pages = str(result["volumeInfo"]["pageCount"])
genre = str(result["volumeInfo"]["categories"])[2:-2]
language = str(result["volumeInfo"]["language"])
image_link = str(result["volumeInfo"]["imageLinks"]["thumbnail"])
dict = Google_Results(isbn13, isbn10, title, author, publisher, published_date, description, pages, genre,
language, image_link)
gr.append(dict)
print(gr[i].title)
i += 1
except:
pass
return
gr = []
Book_Search("Linkedin")
I am a beginner to Python, so any help would be appreciated!
It does so because there is no publisher entry in volumeInfo of the first entry, thus it raises a KeyError and your except captures it. If you're going to work with fuzzy data you have to account for the fact that it will not always have the expected structure. For simple cases you can rely on dict.get() and its default argument to return a 'valid' default entry if an entry is missing.
Also, there are a few conceptual problems with your function - it relies on a global gr which is bad design, it shadows the built-in dict type and it captures all exceptions guaranteeing that you cannot exit your code even with a SIGINT... I'd suggest you to convert it to something a bit more sane:
def book_search(search_term, max_results=3):
results = [] # a list to store the results
parms = {"q": search_term, "maxResults": max_results}
r = requests.get(url="https://www.googleapis.com/books/v1/volumes", params=parms)
try: # just in case the server doesn't return valid JSON
for result in r.json().get("items", []):
if "volumeInfo" not in result: # invalid entry - missing volumeInfo
continue
result_dict = {} # a dictionary to store our discovered fields
result = result["volumeInfo"] # all the data we're interested is in volumeInfo
isbns = result.get("industryIdentifiers", None) # capture ISBNs
if isinstance(isbns, list) and isbns:
for i, t in enumerate(("isbn10", "isbn13")):
if len(isbns) > i and isinstance(isbns[i], dict):
result_dict[t] = isbns[i].get("identifier", None)
result_dict["title"] = result.get("title", None)
authors = result.get("authors", None) # capture authors
if isinstance(authors, list) and len(authors) > 2: # you're slicing from 2
result_dict["author"] = str(authors[2:-2])
result_dict["publisher"] = result.get("publisher", None)
result_dict["published_date"] = result.get("publishedDate", None)
result_dict["description"] = result.get("description", None)
result_dict["pages"] = result.get("pageCount", None)
genres = result.get("authors", None) # capture genres
if isinstance(genres, list) and len(genres) > 2: # since you're slicing from 2
result_dict["genre"] = str(genres[2:-2])
result_dict["language"] = result.get("language", None)
result_dict["image_link"] = result.get("imageLinks", {}).get("thumbnail", None)
# make sure Google_Results accepts keyword arguments like title, author...
# and make them optional as they might not be in the returned result
gr = Google_Results(**result_dict)
results.append(gr) # add it to the results list
except ValueError:
return None # invalid response returned, you may raise an error instead
return results # return the results
Then you can easily retrieve as much info as possible for a term:
gr = book_search("Google")
And it will be far more tolerant of data omissions, provided that your Google_Results type makes most of the entries optional.
Following #Coldspeed's recommendation it became clear that missing information in the JSON file caused the exception to run. Since I only had a "pass" statement there it skipped the entire result. Therefore I will have to adapt the "Try and Except" statements so errors do get handled properly.
Thanks for the help guys!
I'm writing a python scraper code for OpenData and I have one question about : how to check if all values aren't filled in site and if it is null change value to null.
My scraper is here.
Currently I'm working on it to optimalize.
My variables now look like:
evcisloval = soup.find_all('td')[3].text.strip()
prinalezival = soup.find_all('td')[5].text.strip()
popisfaplnenia = soup.find_all('td')[7].text.replace('\"', '')
hodnotafaplnenia = soup.find_all('td')[9].text[:-1].replace(",", ".").replace(" ", "")
datumdfa = soup.find_all('td')[11].text
datumzfa = soup.find_all('td')[13].text
formazaplatenia = soup.find_all('td')[15].text
obchmenonazov = soup.find_all('td')[17].text
sidlofirmy = soup.find_all('td')[19].text
pravnaforma = soup.find_all('td')[21].text
sudregistracie = soup.find_all('td')[23].text
ico = soup.find_all('td')[25].text
dic = soup.find_all('td')[27].text
cislouctu = soup.find_all('td')[29].text
And Output :
scraperwiki.sqlite.save(unique_keys=["invoice_id"],
data={ "invoice_id":number,
"invoice_price":hodnotafaplnenia,
"evidence_no":evcisloval,
"paired_with":prinalezival,
"invoice_desc":popisfaplnenia,
"date_received":datumdfa,
"date_payment":datumzfa,
"pay_form":formazaplatenia,
"trade_name":obchmenonazov,
"trade_form":pravnaforma,
"company_location":sidlofirmy,
"court":sudregistracie,
"ico":ico,
"dic":dic,
"accout_no":cislouctu,
"invoice_attachment":urlfa,
"invoice_url":url})
I googled it but without success.
First, write a configuration dict of your variables in the form:
conf = {'evidence_no': (3, str.strip),
'trade_form': (21, None),
...}
i.e. key is the output key, value is a tuple of id from soup.find_all('td') and of an optional function that has to be applied to the result, None otherwise. You don't need those Slavic variable names that may confuse other SO members.
Then iterate over conf and fill the data dict.
Also, run soup.find_all('td') before the loop.
tds = soup.find_all('td')
data = {}
for name, (num, func) in conf.iteritems():
text = tds[num].text
# replace text with None or "NULL" or whatever if needed
...
if func is None:
data[name] = text
else:
data[name] = func(text)
This will remove a lot of duplicated code. Easier to maintain.
Also, I am not sure the strings "NULL" are the best way to write missing data. Doesn't sqlite support Python's real None objects?
Just read your attached link, and it seems what you want is
evcisloval = soup.find_all('td')[3].text.strip() or "NULL"
But be careful. You should only do this with strings. If the part before or is either empty or False or None, or 0, they will all be replaced with "NULL"