Thread by @Jeande_d

Kaggle's 2021 State of Data Science and Machine Learning survey was released a few days ago.

If you didn't see it, here are some important takeaways 🧵

Top 5 IDEs

1. Jupyter Notebook
2. Visual Studio Code
3. JupyterLab
4. PyCharm
5. RStudio

ML Algorithms Usage: Top 10

1. Linear/logistic regression
2. Decision trees/random forests
3. Gradient boosting machines(Xgboost, LightGBM)
5. Convnets
6. Bayesian approaches
7. Dense neural networks(MLPs)
8. Recurrent neural networks(RNNs)
9. Transformers(BERT, GPT-3)
10. GANs

Machine Learning Tools Landscape - Top 8

1. Scikit-Learn
2. TensorFlow(tf.keras included)
3. XGBoost
4. Keras
5. PyTorch
6. LightGBM
7. CatBoost
8. Huggingface🤗

Cloud Computing Tools - Top 3

1. AWS
2. GCP
3. Microsoft Azure

Enteprise ML Tools - Top 5

1. Amazon SageMaker
2. DataBricks
3. Asure ML Studio
4. Google Cloud Vertex AI
5. DataRobot

Notes: If you look at the graph, it seems that over half the number of the survey responders don't use those kinds of tools.

Databases - Top 4

1. MySQL
2. PostgreSQL
3. Microsoft SQL Server
4. MongoDB

Machine Learning Experimentation Tools - Top 4

1. TensorBoard
2. MLflow
3. Weights & Biases
4. Neptune.ai

Notes: Looking at the graph, the majority of Kagglers do not track their ML models. All eye on the leaderboard!

AutoML Tools - Top 5

1. Google Cloud AutoML
2. Azure Automated ML
3. Amazon SageMaker Autopilot
4. H20 Driverless AI
5. Databricks AutoML

CONCLUSIONS:

1. Notebooks are still the most appreciated way of experimenting with ML. If you never did it, try them in VSCode.
2. Scikit-Learn is ahead of the game
3. All you need is XGBoost(CC: @tunguz)
4. No need for model tracking on Kaggle. There is a leaderboard

If you would like to read the whole survey, here is the link:

kaggle.com/kaggle-survey-…

Thanks for reading.

For more machine learning content, follow @Jeande_d.