John Andrews’ Post

Hi , In my current project on automatic essay scoring, I've implemented a classification approach utilizing BERT. The goal is to assign essays a score ranging from 1 to 6. However, I've encountered a challenge where data points are mispredicted by neighboring classes, such as predicting a score of 3 when the actual score is 2 or 4, occur frequently. As a result, the overall metrics reflect poorly on the model's performance. Upon further investigation, I've identified class imbalance as a significant factor influencing these misclassifications. Addressing this issue could potentially improve the model's performance. However, it's worth noting that despite these misclassifications, when considering the top-3 predictions, over 98% of the data points are correctly classified. While tackling class imbalance is a priority, I'm also open to exploring additional strategies to enhance the model's performance. Any suggestions or insights would be greatly appreciated. Thank you for your input! Code Link : https://lnkd.in/ed_JKHa5 #connection #kaggle #machinelearning #nlp #ai

Mistral AI gives us one of the best foundation model out there. Even with the small parameters, Mistral-7B is able to compete with many company-driven foundation models, such as Llama 2 13B, on all benchmarks. In this article, I want to discuss how to perform LLM foundation model fine-tuning with HuggingFace AutoTrain. With AutoTrain, you don't need to think much about the code complexity and only need to focus on what's important. How does it work? Read it all here! #data #datascience #machinelearning #LLM

Hey, everyone! 👋  Sharing a recent project milestone from my journey at #OasisInfobyte! In Task 1, I tackled the evergreen Iris dataset, implementing Decision Trees. The result? A solid 98% accuracy in predicting iris species. Not bad for predicting which flower is fancier than the others, right?  Here's the scoop: 📦 Data Collection: Utilized Python and Seaborn for dataset handling. 🌿 Feature Engineering: Splitting data into training and testing sets for accuracy assessment. 🌳 Decision Tree Model: Employed Decision Tree Classifier with post-pruning techniques. 🖼️ Visualizing the Decision Tree: Visualized the decision-making process. 🔮 Predictions and Evaluation: Achieved an impressive 98% accuracy on test data. 📊 Classification Report: A closer look at precision, recall, and F1-score. Kudos to #OasisInfobyte for fostering innovation in AI and data science. Excited to dive deeper into the AI realm! #IrisClassification

In this video, we dive into how to run ML and NLP operations in a data processing pipeline at scale. In the first part, we employ Dataflow ML for a well-known ML-NLP application called word clustering. Here, we handle the spaCy and scikit-learn models sequentially in a Vertex AI user-managed notebook for creating four BIRCH clusters for the 300-dimensional word embedding vectors.

In this video, we dive into how to run ML and NLP operations in a data processing pipeline at scale. In the first part, we employ Dataflow ML for a well-known ML-NLP application called word clustering. Here, we handle the spaCy and scikit-learn models sequentially in a Vertex AI user-managed notebook for creating four BIRCH clusters for the 300-dimensional word embedding vectors.