project_work

Projects - for assessing my skills - for data science entry roles

Most of the topics are part of AppliedAI(online course) assignments. I have done some additional work in some assignments. All the implementation are mine own and not plagiarized.

1. cross_attention_bahdanau_attention_implementation: Implementing from scratch bahdanau attention mechanism for language translation. Implemented Encoder, Decoder, Cross-attention layers.
2. eda_chisquare_confidenceinterval_haberman_dataset: EDA on haberman cancer dataset with chi-square test and confidence interval for giving support to observations made through several plots.
3. segmentation_road_traffic: Segmenting Road Traffic using unet. Video: https://drive.google.com/file/d/1Ita9Pe53zS6KC6khZ4rmfw1GvLoq8g3N/view?usp=sharing
4. xgboost_donors_choose: Applying Xgboost to donors choose dataset. Made various Custom Sklearn Transformers having fit/transform methods for easy integration into pipeline(for featurization like response coding, tf-idf weighted w2v encoding, sentiment score).
5. masked_attention_mechanism: Implementing Masked Attention Mechanism as seen in gpt/transformer. attempting to test on ecommerce sessions data.
6. decision_tree_donors_choose: Decision Tree algorithm on donors_choose dataset. Made various Custom Sklearn Transformers having fit/transform methods for easy integration into pipeline(for featurization like response coding, tf-idf weighted w2v encoding, sentiment score).
7. naive_bayes_donors_choose: Naive Bayes on donors_choose dataset
8. sgd_regressor_implementation: Implementing SGD Regressor from scratch
9. tfidf_implementation: Implementing TF-IDF from scratch
10. lstm_donors_choose: LSTM on donors choose dataset
11. conv1d_text_document_classification: applying Conv1D on text for text document classification
12. sql_imdb: SQL on imdb dataset
13. k_fold_cv_knn_implementation: Implementing k fold CV from scratch.
14. densenet_cifar10: Experimenting training with Learning Rate Finder, Cyclical Learning Rate on DenseNet using Cifar10 dataset with keeping no. of parameters less than 1M.