I have fine-tuned one of the pre-trained Bert model and I have set up the maximum sequence length to 500 token. The fine-tuning was successful and the results was quite good. However, when I try to test the fine-tuned model with a sequence length of more than 300 tokens I got the error "index out of range in self".
Why testing the fine-tuned model work only for sequences which is less than 300 tokens in length?
here you see where the maximum sequence length is applied:
.
.
MAX_LEN = 500
input_ids = pad_sequences([tokenizer.convert_tokens_to_ids(txt) for txt in tokenized_texts],
maxlen=MAX_LEN, dtype="long", value=0.0, truncating="post", padding="post")
input_ids = pad_sequences([tokenizer.convert_tokens_to_ids(txt) for txt in tokenized_texts],
maxlen=MAX_LEN, dtype="long", value=0.0, truncating="post", padding="post")
.
.
Related
Say, I have three sample sentences:
s0 = "This model was pretrained using a specific normalization pipeline available here!"
s1 = "Thank to all the people around,"
s2 = "Bengali Mask Language Model for Bengali Language"
I could make a batch like:
batch = [[s[0], s[1]], [s[1], s[2]]]
Now, if I apply the BERT tokenizer on the sentence pairs, it truncates the sentence pairs if the length exceeds in such a way that the ultimate sum of the sentence pairs' lengths meets the max_length parameter, which was supposed to be done, okay. Here's what I meant:
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
model = AutoModelForPreTraining.from_pretrained("bert-base-uncased")
encoded = tokenizer(batch, padding="max_length", truncation=True, max_length=10)["input_ids"]
decoded = tokenizer.batch_decode(encoded)
print(decoded)
>>>Output: ['[CLS] this model was pre [SEP] thank to all [SEP]', '[CLS] thank to all [SEP] bengali mask language model [SEP]']
My question is, how does the truncation work here in the pair of sentences where the number of tokens from each sentence of each pair is not equal?
For example, in the first example output '[CLS] this model was pre [SEP] thank to all [SEP]' number of tokens from the two sentences has not come equally i.e [CLS] 4 tokens [SEP] 3 tokens [SEP].
There are different truncation strategies you can choose from:
True or 'longest_first': Truncate to a maximum length specified with the argument max_length or to the maximum acceptable input length for the model if that argument is not provided. This will truncate token by token, removing a token from the longest sequence in the pair if a pair of sequences (or a batch of pairs) is provided.
'only_first': Truncate to a maximum length specified with the argument max_length or to the maximum acceptable input length for the model if that argument is not provided. This will only truncate the first sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
'only_second': Truncate to a maximum length specified with the argument max_length or to the maximum acceptable input length for the model if that argument is not provided. This will only truncate the second sequence of a pair if a pair of sequences (or a batch of pairs) is provided.
False or 'do_not_truncate' (default): No truncation (i.e., can output batch with sequence lengths greater than the model maximum admissible input size).
I'm running the following using the huggingface implementation:
t1 = "My example sentence is really great."
tokenizer = TransfoXLTokenizer.from_pretrained('transfo-xl-wt103')
model = TransfoXLLMHeadModel.from_pretrained("transfo-xl-wt103")
encoded_input = tokenizer(t1, return_tensors='pt', add_space_before_punct_symbol=True)
output = model(**encoded_input)
tmp = output[0].detach().numpy()
print(tmp.shape)
>>> (1, 7, 267735)
With the goal of getting output embeddings that I'll use downstream.
The last dimension is /substantially/ larger than I expected, and it looks like it is the size of the entire vocab_size rather than a reduction based on the ECL from the paper (which potentially I am misinterpreting).
What argument would I provide the model to reduce this layer size to a smaller dimensional space, something more like the basic BERT at 400 or 768 and still obtain good performance based on the pretrained embeddings?
That's because you used ...LMHeadModel, which predicts the next token. You can use TransfoXLModel.from_pretrained("transfo-xl-wt103") instead, then output[0] is the last hidden state which has the shape (batch_size, sequence_length, hidden_size).
I'm trying to split a dataset into train and test groups in Python using a method similar to what I'm used to in R (I realize there are other options). So I'm defining an array of row numbers that will make up my train set. I then want to grab the remaining row numbers for my test set using np.delete. Since there are 170 rows total and 136 go to the train set, the test set should have 34 rows. But it's got 80 -- the actual number varies when I change my random seed ... What have I got wrong here?
np.random.seed(222)
marriage = np.random.rand(170,55)
rows,cols = marriage.shape
sample = np.random.randint(0,rows-1,(round(.8*rows)))
train = marriage[sample,:]
test = np.delete(marriage, sample, axis=0)
print(marriage.shape)
print(len(sample))
print(train.shape)
print(test.shape)
Is there a way to get the topic distribution of an unseen document using a pretrained LDA model without using the LDA_Model[unseenDoc] syntax? I am trying to implement my LDA model into a web application, and if there was a way to use matrix multiplication to get a similar result then I could use the model in javascript.
For example, I tried the following:
import numpy as np
import gensim
from gensim.corpora import Dictionary
from gensim import models
import nltk
from nltk.stem import WordNetLemmatizer, SnowballStemmer
nltk.download('wordnet')
def Preprocesser(text_list):
smallestWordSize = 3
processedList = []
for token in gensim.utils.simple_preprocess(text_list):
if token not in gensim.parsing.preprocessing.STOPWORDS and len(token) > smallestWordSize:
processedList.append(StemmAndLemmatize(token))
return processedList
lda_model = models.LdaModel.load('LDAModel\GoldModel') #Load pretrained LDA model
dictionary = Dictionary.load("ModelTrain\ManDict") #Load dictionary model was trained on
#Sample Unseen Doc to Analyze
doc = "I am going to write a string about how I can't get my task executor \
to travel properly. I am trying to use the \
AGV navigator, but it doesn't seem to be working network. I have been trying\
to use the AGV Process flow but that isn't working either speed\
trailer offset I am now going to change this so I can see how fast it runs"
termTopicMatrix = lda_model.get_topics() #Get Term-topic Matrix from pretrained LDA model
cleanDoc = Preprocesser(doc) #Tokenize, lemmatize, clean and stem words
bowDoc = dictionary.doc2bow(cleanDoc) #Create bow using dictionary
dictSize = len(termTopicMatrix[0]) #Get length of terms in dictionary
fullDict = np.zeros(dictSize) #Initialize array which is length of dictionary size
First = [first[0] for first in bowDoc] #Get index of terms in bag of words
Second = [second[1] for second in bowDoc] #Get frequency of term in bag of words
fullDict[First] = Second #Add word frequency to full dictionary
print('Matrix Multiplication: \n', np.dot(termTopicMatrix,fullDict))
print('Conventional Syntax: \n', lda_model[bowDoc])
Output:
Matrix Multiplication:
[0.0283254 0.01574513 0.03669142 0.01671816 0.03742738 0.01989461
0.01558603 0.0370233 0.04648389 0.02887623 0.00776652 0.02147539
0.10045133 0.01084273 0.01229849 0.00743788 0.03747379 0.00345913
0.03086953 0.00628912 0.29406082 0.10656977 0.00618827 0.00406316
0.08775404 0.00785408 0.02722744 0.09957815 0.01669402 0.00744392
0.31177135 0.03063149 0.07211428 0.01192056 0.03228589]
Conventional Syntax:
[(0, 0.070313625), (2, 0.056414187), (18, 0.2016589), (20, 0.46500313), (24, 0.1589748)]
In the pretrained model there are 35 topics and 1155 words.
In the "Conventional Syntax" output, the first element of each tuple is the index of the topic and the second element is the probability of the topic. In the "Matrix Multiplication" version, the probability is the index and the value is the probability. Clearly the two don't match up.
For example, the lda_model[unseenDoc] shows that topic 0 has a 0.07 probability, but the matrix multiplication method says that topic has a 0.028 probability. Am I missing a step here?
You can review the full source code used by LDAModel's get_document_topics() method in your installation, or online at:
https://github.com/RaRe-Technologies/gensim/blob/e75f6c8e8d1dee0786b1b2cd5ef60da2e290f489/gensim/models/ldamodel.py#L1283
(It also makes use of the inference() method in the same file.)
It's doing a lot more scaling/normalization/clipping than your code, which is likely the cause of the discrepancy. But you should be able to examine, line-by-line, where your process & its differ to get the steps to match up.
It also shouldn't be hard to use the gensim code's steps as guidance for creating parallel Javascript code that, given the right parts of the model's state, can reproduce its results.
Here is the problem, where I need to transform an ID (defined as a long integer) to a smaller alfanumeric identifier. The details are the following:
Each individual on the problem as an unique ID, a long integer of size 13 (something like 123123412341234).
I need to generate a smaller representation of this unique ID, a alfanumeric string, something like A1CB3X. The problem is that 5 or 6 character length will not be enough to represent such a large integer.
The new ID (eg A1CB3X) should be valid in a context where we know that only a small number of individuals are present (less than 500). The new ID should be unique within that small set of individuals.
The new ID (eg A1CB3X) should be the result of a calculation made over the original ID. This means that taking the original ID elsewhere and applying the same calculation, we should get the same new ID (eg A1CB3X).
This calculation should occur when the individual is added to the set, meaning that not all individuals belonging to that set will be know at that time.
Any directions on how to solve such a problem?
Assuming that you don't need a formula that goes in both directions (which is impossible if you are reducing a 13-digit number to a 5 or 6-character alphanum string):
If you can have up to 6 alphanumeric characters that gives you 366 = 2,176,782,336 possibilities, assuming only numbers and uppercase letters.
To map your larger 13-digit number onto this space, you can take a modulo of some prime number slightly smaller than that, for example 2,176,782,317, the encode it with base-36 encoding.
alphanum_id = base36encode(longnumber_id % 2176782317)
For a set of 500, this gives you a
2176782317P500 / 2176782317500 chance of a collision
(P is permutation)
Best option is to change the base to 62 using case sensitive characters
If you want it to be shorter, you can add unicode characters. See below.
Here is javascript code for you: https://jsfiddle.net/vewmdt85/1/
function compress(n) {
var symbols = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyzÀÁÂÃÄÅÆÇÈÉÊËÌÍÎÏÐÑÒÓÔÕÖØÙÚÛÜÝÞßàáâãäåæçèéêëìíîïð'.split('');
var d = n;
var compressed = '';
while (d >= 1) {
compressed = symbols[(d - (symbols.length * Math.floor(d / symbols.length)))] + compressed;
d = Math.floor(d / symbols.length);
}
return compressed;
}
$('input').keyup(function() {
$('span').html(compress($(this).val()))
})
$('span').html(compress($('input').val()))
How about using some base-X conversion, for example 123123412341234 becomes 17N644R7CI in base-36 and 9999999999999 becomes 3JLXPT2PR?
If you need a mapping that works both directions, you can simply go for a larger base.
Meaning: using base 16, you can reduce 1 to 16 to a single character.
So, base36 is the "maximum" that allows for shorter strings (when 1-1 mapping is required)!