#DSFthegreatindoors presents

Workshop - Text Classification by Transfer learning with Deep Transformers - Depop @depop
A talk by Oduwa Edo-Osagie @odieED, Data Scientist, Depop

#TextMining
#NLP
#AI
#DataScience
#MachineLearning
#DataEngineering
@DataIdols
#DSFthegreatindoors
ND Image
About the talk

Workshop - Text Classification by Transfer learning with Deep Transformers - Depop @depop
A talk by Oduwa Edo-Osagie @odieED, Data Scientist, Depop

#TextMining
#NLP
#AI
#DataScience
#MachineLearning
#DataEngineering

@DataIdols

#DSFthegreatindoors
ND ImageImage
1st learn features, then work on the classifier
Text Classification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#MachineLearning
#DataEngineering
@DataIdols
#DSFthegreatindoors
ND Image
Strategies for transfer learning?
1st learn features, then work on the classifier
Te#xtClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop
#TextMining
#NLP
#AI #CNN
#DataScience
#ML
@DataIdols
#DSFthegreatindoors Image
Crowdsourced image labelling that's gamified is a good approach
#TextClassification by #TransferLearning w/ #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop
#TextMining
#NLP
#AI #CNN
#DataScience
#ML
#DataEngineering
@DataIdols
#DSFthegreatindoors
ND ImageImage
Batch normalisation helps reduce training time

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
BERT Bidirectional Encoder Representations from Transformers
#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#ML
#DataEngineering
@DataIdols
#DSFthegreatindoors
ND Image
BERT has 12 transformer layers & 110 million parameters
Training BERT

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#ML
@DataIdols
#DSFthegreatindoors
ND Image
Notice word piece embeddings (in image) to save memory

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
Time for some Q&A before we head to the hands-on session

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND ImageImage
Notice line 64 onwards
input_ids input_mask segment_ids
#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND ImageImage
Line 24 call the BertLayer, notice the kwargs

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
Now train the last 10 layers line 33 onwards

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
Now we build our own BERT
Create a BERT class

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
Now the masks and segements line 25 onwards

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
BERT does feature extraction line 31 onwards

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
#DataEngineering

@DataIdols
#DSFthegreatindoors
ND Image
We get vectors for CLS i sep
Line 31 Get matrix 768 x 64 x 4
#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND Image
Now for a visual representation
Use loss=binary_crossentropy L 36
#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop
#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
@DataIdols
#DSFthegreatindoors
ND Image
Line 41 Use 90% of data for training, 10% for testing

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND ImageImage
Line 52 Ready to train the model with 2 epochs

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND Image
Line 56 Set batch_size-32, this is another hyper-parameter

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND ImageImage
Line 25 We need to initialise the session

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND Image
Line 67 initialise BERT with data CSV in git repo (dim 500 x 2)

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND ImageImage
Pls RETWEET, SHARE & MENTION 🙏
Pls RETWEET, SHARE & MENTION 🙏

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop
#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML
@DataIdols
#DSFthegreatindoors
ND ImageImageImageImage
Note
val_acc 0.82 (validation accuracy)
Total params 110,302,011

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop
#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#MLS
@DataIdols
#DSFthegreatindoors
ND ImageImage
Git repo updated in real-time by Oduwa! Thanks.

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND ImageImage
Line 65 Now let's use the model to make spam PREDICTIONS
#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND Image
Notice 0.78 probability the dummy data is spam

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND Image
Depop are hiring for a Jnr Data Scientist! Get in touch!

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie @odieED Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND ImageImageImage
MASSIVE THANKS TO
Oduwa Edo-Osagie @odieED & @depop

Pls RETWEET 🙏🙏

#TextClassification by #TransferLearning with #DeepTransformers @depop
By Oduwa Edo-Osagie Snr Data Scientist @depop

#TextMining
#NLP
#TensorFlow
#AI #CNN
#DataScience
#ML

@DataIdols
#DSFthegreatindoors
ND ImageImageImageImage

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Data Science Festival

Data Science Festival Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @DataScienceFest

28 Nov
#DSFthegreatindoors presents

Workshop
Going postal: how to craft a cutting-edge route optimisation engine in-house that suits your business needs
A talk by Fabrice Durier, Hugo Galy and Louisa Sober of @RoyalMail

#DataScience
#MachineLearning
#optimisation

@DataIdols
ND Image
We are off

Workshop
Going postal: how to craft a cutting-edge route optimisation engine in-house that suits your business needs
A talk by Fabrice Durier, Hugo Galy and Louisa Sober of @RoyalMail

#DataScience
#MachineLearning
#optimisation

@DataIdols
#DSFthegreatindoors
ND ImageImage
The diverse team
Workshop
Going postal: how to craft a cutting-edge route optimisation engine in-house that suits your business needs
A talk by Fabrice Durier, Hugo Galy and Louisa Sober of @RoyalMail

#DataScience
#MachineLearning
#optimisation
@DataIdols
#DSFthegreatindoors
ND ImageImage
Read 10 tweets
26 Nov
Panel Session

#ClimateChange: Can Tech save the world?

A talk by Vijay Bhopal @Vij_scene, John Booth @Carbon3IT, Lubomila Jordanova @LubomilaJ, Brittany Salas @brittanymsalas & Laurence Watson

@DataIdols
#DSFthegreatindoors
ND Image
We are off....

Panel Session

#ClimateChange: Can Tech save the world?

A talk by Vijay Bhopal @Vij_scene, John Booth @Carbon3IT, Lubomila Jordanova @LubomilaJ, Brittany Salas @brittanymsalas & Laurence Watson

@DataIdols
#DSFthegreatindoors
ND ImageImage
Great to have host: Vijay Bhopal @Vij_scene

Panel Session

#ClimateChange: Can Tech save the world?

A talk by Vijay Bhopal @Vij_scene, John Booth @Carbon3IT, Lubomila Jordanova @LubomilaJ, Brittany Salas @brittanymsalas & Laurence Watson

@DataIdols
#DSFthegreatindoors
ND Image
Read 43 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!