Extract probabilities and labels from FARM TextClassification - python

I have spent a few days exploring the excellent FARM library and its modular approach to building models. The default output (result) however is very verbose, including a multiplicity of texts, values and ASCII artwork. For my research I only require the predicted labels from my NLP text classification model, together with the individual probabilities. How do I do that? I have been experimenting with nested lists/dictionaries but am unable to neatly produce a simple list of output labels and probabilities.
enter code here
# Test your model on a sample (Inference)
from farm.infer import Inferencer
from pprint import PrettyPrinter
infer_model = Inferencer(processor=processor, model=model, task_type="text_classification", gpu=True)
basic_texts = [
# a snippet or two from Dickens
{"text": "Mr Dombey had remained in his own apartment since the death of his wife, absorbed in visions of the youth, education, and destination of his baby son. Something lay at the bottom of his cool heart, colder and heavier than its ordinary load; but it was more a sense of the child’s loss than his own, awakening within him an almost angry sorrow."},
{"text": "Soon after seven o'clock we went down to dinner, carefully, by Mrs. Jellyby's advice, for the stair-carpets, besides being very deficient in stair-wires, were so torn as to be absolute traps."},
{"text": "Walter passed out at the door, and was about to close it after him, when, hearing the voices of the brothers again, and also the mention of his own name, he stood irresolutely, with his hand upon the lock, and the door ajar, uncertain whether to return or go away."},
# from Lewis Carroll
{"text": "I have kept one for many years, and have found it of the greatest possible service, in many ways: it secures my _answering_ Letters, however long they have to wait; it enables me to refer, for my own guidance, to the details of previous correspondence, though the actual Letters may have been destroyed long ago;"},
{"text": "The Queen gasped, and sat down: the rapid journey through the air had quite taken away her breath and for a minute or two she could do nothing but hug the little Lily in silence."},
{"text": "Rub as she could, she could make nothing more of it: she was in a little dark shop, leaning with her elbows on the counter, and opposite to her was an old Sheep, sitting in an arm-chair knitting, and every now and then leaving off to look at her through a great pair of spectacles."},
# G K Chesterton
{"text": "Basil and I walked rapidly to the window which looked out on the garden. It was a small and somewhat smug suburban garden; the flower beds a little too neat and like the pattern of a coloured carpet; but on this shining and opulent summer day even they had the exuberance of something natural, I had almost said tropical. "},
{"text": "This is the whole danger of our time. There is a difference between the oppression which has been too common in the past and the oppression which seems only too probable in the future."},
{"text": "But whatever else the worst doctrine of depravity may have been, it was a product of spiritual conviction; it had nothing to do with remote physical origins. Men thought mankind wicked because they felt wicked themselves. "},
]
result = infer_model.inference_from_dicts(dicts=basic_texts)
PrettyPrinter().pprint(result)
#print(result)

All logging (incl. the ASCII artwork) is done in FARM via Python's logging framework. You can simply disable the logs up to a certain level like this at the beginning of your script:
import logging
logging.disable(logging.ERROR)
Is that what you are looking for or do you rather want to adjust the output format of the model predictions? If you only need label and probability, you could do something like this:
...
basic_texts = [
{"text": "Stackoverflow is a great community"},
{"text": "It's snowing"},
]
infer_model = Inferencer(processor=processor, model=model, task_type="text_classification", gpu=True)
result = infer_model.inference_from_dicts(dicts=basic_texts)
minimal_results = []
for sample in result:
# Only extract the top 1 prediction per sample
top_pred = sample["predictions"][0]
minimal_results.append({"label": top_pred["label"], "probability": top_pred["probability"]})
PrettyPrinter().pprint(minimal_results)
infer_model.close_multiprocessing_pool()
(I left out the initial model loading etc. - see this example for more details)

Related

Python package to extract sentence from a textfile based on keyword

I need a python package that could get the related sentence from a text, based on the keywords provided.
For example, below is the Wikipedia page of J.J Oppenheimer -
Early life
Childhood and education
J. Robert Oppenheimer was born in New York City on April 22, 1904,[note 1][7] to Julius Oppenheimer, a wealthy Jewish textile importer who had immigrated to the United States from Germany in 1888, and Ella Friedman, a painter.
Julius came to the United States with no money, no baccalaureate studies, and no knowledge of the English language. He got a job in a textile company and within a decade was an executive with the company. Ella was from Baltimore.[8] The Oppenheimer were non-observant Ashkenazi Jews.[9]
The first atomic bomb was successfully detonated on July 16, 1945, in the Trinity test in New Mexico.
Oppenheimer later remarked that it brought to mind words from the Bhagavad Gita: "Now I am become Death, the destroyer of worlds.
If my passed string is - "JJ Oppenheimer birth date", it should return "J. Robert Oppenheimer was born in New York City on April 22, 1904"
If my passed string is - "JJ Openheimer Trinity test", it should return "The first atomic bomb was successfully detonated on July 16, 1945, in the Trinity test in New Mexico"
I tried searching a lot but nothing comes closer to what I want and I don't know much about NLP vectorization techniques. It would be great if someone please suggest some package if they know(or exist).
You could use fuzzywuzzy.
fuzz.ratio(search_text, sentence).
This gives you a score of how similar two strings are.
https://github.com/seatgeek/fuzzywuzzy
I am pretty sure a Module exists that could do this for you, you could try and make it yourself by parsing through the text and creating words like: ["date of birth", "born", "birth date", etc] and you do this for multiple fields. This would thus allow you to find information that would be available.
The idea is:
you grab your text or whatever u have,
you grab what you are looking for (example date of birth)
You then assign a date of birth to a list of similar words,
you look through ur file to see if you find a sentence that has that in it.
I am pretty sure there is no module, maybe I am wrong but smth like this should work.
The task you describe looks like Information Retrieval. Given a query (the keywords) the model should return a list of document (the sentences) that best matches the query.
This is essentially what the response using fuzzywuzzy is suggesting. But maybe just counting the number of occurences of the query words in each sentence is enough (and more efficient).
The next step would be to use Tf-Idf. It is a weighting scheme, that gives high scores to words that are specific to a document with respect to a set of document (the corpus).
This results in every document having a vector associated, you will then be able to sort the documents according to their similarity to the query vector. SO Answer to do that

Working with span objects. [spaCy, python]

I'm not sure if this is a really dumb question, but here goes.
text_corpus = '''Insurance bosses plead guilty\n\nAnother three US insurance executives have pleaded guilty to fraud charges stemming from an ongoing investigation into industry malpractice.\n\nTwo executives from American International Group (AIG) and one from Marsh & McLennan were the latest. The investigation by New York attorney general Eliot Spitzer has now obtained nine guilty pleas. The highest ranking executive pleading guilty on Tuesday was former Marsh senior vice president Joshua Bewlay.\n\nHe admitted one felony count of scheming to defraud and faces up to four years in prison. A Marsh spokeswoman said Mr Bewlay was no longer with the company. Mr Spitzer\'s investigation of the US insurance industry looked at whether companies rigged bids and fixed prices. Last month Marsh agreed to pay $850m (£415m) to settle a lawsuit filed by Mr Spitzer, but under the settlement it "neither admits nor denies the allegations".\n'''
def get_entities(document_text, model):
analyzed_doc = model(document_text)
entities = [entity for entity in analyzed_doc.ents if entity.label_ in ["PER", "ORG", "LOC", "GPE"]]
return entities
model = spacy.load("en_core_web_sm")
entities_1 = get_entities(text_corpus, model)
entities_2 = get_entities(text_corpus, model)
but when it run the following,
entities_1[0] in entities_2
The output is False.
Why is that? The objects in both the entity lists are the same. Yet an item from one list is not in the other one. That's extremely odd. Can someone please explain why that is so to me?
This is due to the way ents's are represented in spaCy. They are classes with specific implementations so even entities_2[0] == entities_1[0] will evaluate to False. By the looks of it, the Span class does not have an implementation of __eq__ which, at first glance at least, is the simple reason why.
If you print out the value of entities_2[0] it will give you US but this is simply because the span class has a __repr__ method implemented in the same file. If you want to do a boolean comparison, one way would be to use the text property of Span and do something like:
entities_1[0].text in [e.text for e in entities_2]
edit:
As #abb pointed out, Span implements __richcmp__, however this is applicable to the same instance of Span since it checks the position of the token itself.

Want to extract text from a text or pdf file as different paragraphs

Check the following text piece
IN THE HIGH COURT OF GUJARAT AT AHMEDABAD
R/CRIMINAL APPEAL NO. 251 of 2009
FOR APPROVAL AND SIGNATURE:
HONOURABLE MR.JUSTICE R.P.DHOLARIA
==========================================================
1 Whether Reporters of Local Papers may be allowed to see the judgment ?
2 To be referred to the Reporter or not ?
3 Whether their Lordships wish to see the fair copy of the judgment ?
4 Whether this case involves a substantial question of law as to the interpretation of the Constitution of India or any order made thereunder ?
========================================================== STATE OF GUJARAT,S M RAO,FOOD INSPECTOR,OFFICE OF THE Versus DHARMESHBHAI NARHARIBHAI GANDHI ========================================================== Appearance: MS HB PUNANI, APP (2) for the Appellant(s) No. 1 MR DK MODI(1317) for the Opponent(s)/Respondent(s) No. 1 ==========================================================
CORAM: HONOURABLE MR.JUSTICE R.P.DHOLARIA
Date : 12/03/2019
ORAL JUDGMENT
1. The appellant State of Gujarat has
preferred the present appeal under section 378(1)
(3) of the Code of Criminal Procedure, 1973
against the judgment and order of acquittal dated
Page 1 of 12
R/CR.A/251/2009 JUDGMENT
17.11.2008 rendered by learned 2nd Additional
Civil Judge and Judicial Magistrate, First Class,
Nadiad in Food Case No.1 of 2007.
The short facts giving rise to the
present appeal are that on 10.11.2006 at about
18.00 hours, the complainant visited the place of
the respondent accused situated at Juna
Makhanpura, Rabarivad, Nadiad along with panch
witness and the respondent was found dealing in
provisional items. The complainant identified
himself as a Food Inspector and after giving
intimation in Form No.6 has purchased muddamal
sample of mustard seeds in the presence of the
panchas for the purpose of analysis. Thereafter,
the complainant Food Inspector has divided the
said sample in equal three parts and after
completing formalities of packing and sealing
obtained signatures of the vendor and panchas and
out of the said three parts, one part was sent to
the Public Analyst, Vadodara for analysis and
remaining two parts were sent to the Local Health
Authority, Gandhinagar. Thereafter, the Public
Analyst forwarded his report. In the said report,
it is stated that the muddamal sample of mustard
seeds is misbranded which is in breach of the
provisions of the Food Adulteration Act, 1954
(for short “the Act”) and the Rules framed
thereunder. It is alleged that, therefore, the
sample of mustard seeds was misbranded and,
thereby, the accused has committed the offence.
**Page 2 of 12
R/CR.A/251/2009* JUDGMENT*
Hence, the complaint came to be lodged against
the respondent accused.
I want to be able to write a program such that it follows the given constraints. Be wary of the fact that this is only a single file i have like 40k files and it should run on all the files. All the files have some difference but the basic format for every file is the same.
Constraints.
It should start the text extraction process from after the "metadata" . Metadata is the data about the file from the starting of the file i.e " In the high court of gujarat" till Oral Judgment. In all the files i have , there are various POINTS after the string ends. So i need all these points as a separate paragraph ( see the text has 2 points , i need it in different paragraphs ).
Check the lines in italics, these are the panes in the text/pdf file. I need to remove these as these donot have any meaning to the text content i want.
These files are both available in TEXT or PDF format so i can use either. But i am new to python so i dont know how and where to start. I just have basic knowledge in python.
This data is going to be made into a "corpus" for further processes in building a huge expert system so you know what needs to be done i hope.
Read the official python docs!
Start with python's basic str type and its methods. One of its methods, find, will find substrings in your text.
Use the python slicing notation to extract the portion of text you need, e.g.
text = """YOUR TEXT HERE..."""
meta_start = 'In the high court of gujarat'
meta_end = 'ORAL JUDGMENT'
pos1 = text.find(meta_start)
pos2 = text.find(meta_end)
if pos2 > pos1 and pos1 > -1:
# text is found, extract it
text1 = text[meta_start + len(meta_start):meta_end - 1]
After that, you can go ahead and save your extracted text to a database.
Of course, a better and more complicated solution would be to use regular expressions, but that's another story -- try finding the right way for yourself!
As to italics and other text formatting, you won't ever be able to mark it out in plain text (unless you have some 'meta' markers, like e.g. [i] tags).

Analysing English text with some French name

I'm dealing with the Well-known novel of Victor Hugo "Les Miserables".
A part of my project is to detect the existence of each of the novel's character in a sentence and count them. This can be done easily by something like this:
def character_frequency(character_per_sentences_dict,):
characters_frequency = OrderedDict([])
for k, v in character_per_sentences_dict.items():
if len(v) != 0:
characters_frequency[k] = len(v)
return characters_frequency, characters_in_vol
This pies of could works well for all of the characters except "Èponine".
I also read the text with the following piece code:
import codecs
import nltk.tokenize
with open(path_to_volume + '.txt', 'r', encoding='latin1') as fp:
novel = ' '.join(fp.readlines())
# Tokenize sentences and calculate the number of sentences
sentences = sent_tokenize(novel)
num_volume = path_to_volume.split("-v")[-1]
I should add that the dictation of "Èponine" is the same everywhere.
Any idea what's going on ?!
Here is a sample in which this name apears:
" ONE SHOULD ALWAYS BEGIN BY ARRESTING THE VICTIMS
At nightfall, Javert had posted his men and had gone into ambush himself between the trees of the Rue de la Bar­rieredes-Gobelins which faced the Gorbeau house, on the other side of the boulevard. He had begun operations by opening his pockets, and dropping into it the two young girls who were charged with keeping a watch on the ap­proaches to the den. But he had only caged Azelma. As for Èponine, she was not at her post, she had disappeared, and he had not been able to seize her. Then Javert had made a point and had bent his ear to waiting for the signal agreed upon. The comings and goings of the fiacres had greatly agi­tated him. At last, he had grown impatient, and, sure that there was a nest there, sure of being in luck, having recog­nized many of the ruffians who had entered, he had finally decided to go upstairs without waiting for the pistol-shot."
I agree with #BoarGules that there is likely a more efficient and effective way to approach this problem. With that said, I'm not sure what your problem is here. Python is fully Unicode supportive. You can "just do it" in terms of using Unicode in your program logic using Python's standard string ops and libraries.
For example, this works:
#!/usr/bin/env python
import requests
names = [
u'Éponine',
u'Cosette'
]
# Retrieve Les Misérables from Project Gutenberg
t = requests.get("http://www.gutenberg.org/files/135/135-0.txt").text
for name in names:
c = t.count(name)
print("{}: {}".format(name, c))
Results:
Éponine: 81
Cosette: 1004
I obviously don't have the text you have, so I don't know if how it is encoded, or how it is being read is the problem. I can't test that without having it. In this code, I get the source text off the internet. My point is just that non-ASCII characters should not pose any impediment to you as long as your inputs are reasonable.
All of the time to run this is spent reading the text. I think even if you added dozens of names, it wouldn't add up to a noticeable delay on any decent computer. So this method works just fine.

Two input, one output Button

I am trying to make a button that takes in the state a person lives in and their race and outputs their life expectancy. I have a long list of data for all the states and various races. I also have my button set up, but need to find a way to use this data in an efficient way that allows the button to take in the two pieces of information and output one number, the life expectancy that correctly corresponds.
Thanks!
Here is what I have so far-
def life_expectancy_race(b):
'''This will tell you your life expectancy based on the information plugged into the boxes.'''
#number = print(text_input1.value) and print(text_input2.value)
display.clear_output()
display.display(display.Latex("Your life expectancy is {} y ears".format(life)))
text_input1= widgets.Text(description='Enter the state you live in here')
text_input2= widgets.Text(description='Enter your race here (White, Native American, Latino, Asian American, African American )')
button = widgets.Button(description='What is my life expectancy?')
button.on_click(life_expectancy_race)
display.display(text_input1)
display.display(text_input2)
display.display(button)
Really not sure what your asking are you asking but would something like this work? I'm not saying that specifically, just whipped that up :/ but something in that general thought process. Sorry if I misunderstood your question.
white = 30
America = 40
Print (Text1 + text2)

Categories

Resources