Comments (3)
It is normal for the source embeddings to move, since the source embeddings are mapped to the space of the target embeddings. So it is expected to have different source embeddings. The target embeddings are not supposed to move though.
from muse.
Hi,
By default, your embeddings on the target side should not move. If they move, it maybe because you are setting --normalize_embeddings
to something like renorm
or center
to renormalize or center your embeddings.
What command are you using?
from muse.
Hi, thanks for the reply, the spaces are definetely moving, I can see in dump folders that the vectors of the source language differs from the original ones:
2519370 300
, -0.02948 -0.21129 0.04343 0.19881 0.05295 -0.10741 0.04260 -0.07466 -0.21796 -0.01356 -0.00595 0.08293 -0.05613 0.13110 0.24169 0.01339 -0.26227 -0.02290 0.03767 -0.$
. 0.06645 -0.29178 0.14053 0.04299 0.13258 -0.22219 -0.03270 -0.01460 0.02280 -0.07971 -0.03245 0.12176 0.00496 -0.10281 0.31369 0.09630 -0.33772 0.14261 0.07439 -0.07$
the 0.04159 -0.11443 0.02923 -0.12095 -0.10245 -0.04293 -0.09272 -0.09758 0.03041 -0.19944 -0.00167 -0.12472 0.01751 -0.11062 0.19063 0.01375 -0.16086 0.02883 0.14020 $
</s> -0.00185 0.00126 0.00013 -0.00325 0.00110 0.00074 -0.00370 0.00007 0.00201 0.00354 0.00259 0.00131 0.00116 -0.00094 0.00011 -0.00213 0.00201 0.00294 -0.00241 -0.0$
of -0.04616 -0.07976 0.06595 -0.10639 -0.00265 -0.02144 -0.04606 -0.04847 -0.00289 -0.23925 0.05506 -0.07128 0.06057 -0.12338 0.23568 0.24877 -0.09040 0.08396 0.06380 $
- 0.08003 -0.35446 0.25302 -0.13906 0.12771 0.05924 -0.23303 -0.12729 -0.13709 -0.22097 0.05848 0.07262 0.18509 0.21290 0.10258 0.13504 -0.05400 0.00911 0.19758 0.3225$
in -0.01183 0.06227 0.03939 0.05250 -0.03320 -0.07291 0.06198 0.02659 -0.17706 -0.03607 -0.01088 -0.09639 -0.03377 -0.10644 0.13871 -0.12476 -0.11023 0.04956 -0.00823 $
and -0.01185 -0.10965 0.09453 0.14128 -0.02225 -0.16218 -0.00361 -0.01935 -0.10978 -0.12176 -0.03559 -0.05627 0.02960 0.02760 0.04719 0.01789 -0.07090 0.02324 0.02459 $
2519370 300
, -0.023167 -0.0042483 -0.10572 0.042783 -0.14316 -0.078954 0.078187 -0.19454 0.022303 0.31207 0.057462 -0.11589 0.096633 -0.093229 -0.034229 -0.14652 -0.11094 -0.1110$
. -0.11112 -0.0013859 -0.1778 0.064508 -0.24037 0.031087 -0.030144 -0.36883 -0.043855 0.24831 0.078633 -0.16072 0.10528 -0.09622 -0.077742 -0.28262 -0.13013 0.0056083 $
the -0.065334 -0.093031 -0.017571 0.20007 0.029521 -0.03992 -0.16328 -0.072946 0.089604 0.080907 -0.040032 -0.23624 0.1825 -0.061241 -0.064386 -0.075258 -0.050076 -0.0$
</s> 0.050258 -0.073228 0.43581 0.17483 -0.18546 -0.39921 -0.50767 -0.5066 -0.15557 0.031451 -0.23794 -0.44625 -0.26341 -0.26413 -0.26935 -0.62865 -0.13574 0.0697 0.26$
of 0.048804 -0.28528 0.018557 0.20577 0.060704 0.085446 -0.036267 -0.068373 0.14507 0.17852 0.14579 -0.1363 0.23348 0.029758 -0.22001 -0.0045515 -0.11197 -0.041367 0.0$
- -0.12278 -0.036748 0.20728 -0.018277 -0.0016348 0.023735 -0.03712 -0.28608 -0.19088 -0.068688 0.061755 -0.052416 0.16867 -0.1108 -0.11308 -0.27392 -0.30827 0.18204 0$
in 0.12367 -0.13965 0.044877 0.18919 -0.10997 -0.0064458 0.050499 -0.20439 -0.015761 0.15049 0.13774 -0.068241 0.17078 -0.13529 -0.18324 -0.00035567 -0.099566 -0.14549$
and -0.031533 0.046278 -0.12534 0.19165 -0.1266 -0.012853 0.10342 -0.0098085 0.15189 0.27582 0.13695 0.0088799 0.14132 -0.12 -0.063439 -0.15178 0.09809 -0.1201 -0.0690$
My command to generate the alignment. My target corpus is very small, just 67 words with 65 in common.
python supervised.py --src_lang en --tgt_lang eg --src_emb ./../x/embeddings/wiki.en.bin --tgt_emb ./../x/embeddings/car_brands.bin --dico_train identical_char --dico_max_rank 0 --n_refinement 5
from muse.
Related Issues (20)
- why unsupervised can achieve Word alignment?
- Can some one give the dictionary tree of the whole project? Like in the data/crosslingual or monlingual/.. HOT 5
- non-parallel chinese traditional - english
- evaluate.py error
- openssl ssl_read ssl_error_syscall errno 110
- Reproducing Results in Table 1 HOT 1
- IndexError: index out of range in self
- AttributeError: 'Namespace' object has no attribute 'dico_max_rank'
- Assertion Error while using the unsupervised way.
- Tokenization issue in to-En bilingual dictionaries
- They hated the kid HOT 1
- Bad outcome in ja-en task HOT 1
- Rush Shhh INPUT aUTOMATION
- ValueError: too many values to unpack (expected 2) in unsupervised.py
- Will pytorch's deprecation of volatile affect the result?
- [ML Question] Is it possible somehow to translate two or three words ?
- Tried on GloVe?
- self-mapped english words in dictionaries
- ValueError: Function has keyword-only parameters or annotations, use inspect.signature() API which can support them HOT 3
- demo notebook references unavailable private files
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from muse.