It's hard to have a clear view on what to learn and what to know to be employable. Especially when you're not in a traditional cursus.

This list is a compilation of most-wanted skills for data scientist based on online job offers.

I took hundreds of data scientist job offers in Paris, France, in Novembre 2020. This list may not be representative of the most-wanted skills in other areas or countries.

The raw data extracted from job offers is visible in JobOffers.md.

The lists are ordered by frequence of mentionning in the offers.

• Understand and implement scientistic papers.
• Statistical methodology. Statistics testing, P-value.

• General statistics knowledge. Distribution, Bayesian inference, statistics models, probabilities.
• Time series analysis.
• Sequential analysis.
• Scoring.
• Regression.
• Econometrics.
• Game theory.

• Complexity estimation.
• Graph theory.
• Approximation algorithm.
• K-nearest neigbours.

• Deep learning. Neural networks theory.
• Decision tree / Gradient boosted decision tree.
• Regression / Logistic regression.
• Reinforcement learning.
• Convolutional Neural Network.
• Neural language processing.
• Ensemble modeling.
• Recommendation.
• Clustering.
• Auto-encoder.
• Restricted Boltzmann machine.

• Qlik.
• Google Data Studio.
• Plotly / Dash. For Python/R.
• Shiny. For R.
• Chartio.
• Matplotlib / Seaborn. For Python.
• Bokeh. For Python, R wrapper.
• Graphiz. For Python/R.
• Kibana.
• PowerBI.
• Sweetviz. For Python.

• Dataiku.
• Druid.
• H2O.ai.

Python was 2x more mentionned than R, but both are really demanded.

SQL is as demanded as R, it appears to be an essential skill.

Dashboarding in general is a top-demanded skill.

• Python.
• R.
• C++.

• Pandas / Numpy. Essential Python data handling libs.
• Scikit-learn.
• Tensorflow / Keras.
• PyTorch.
• PySpark. Connect your Python script to a Spark stack.
• NLTK. Neural language processing lib.
• Scipy.
• MxNet. Deep learning lib.
• XGboost. Gradient boosted decision trees in Pyhton and R.
• Catboost. Yandex boosted gradient decision trees in Python and R.
• LGBM. Microsoft boosted gradient decision trees in Python and R.
• Prophet. Facebook time series forecasting lib.
• Libsvm. Support vector machines in Python.

• Apache Spark. With Hive and AirFlow.
• Hadoop.
• Tableau.
• Linux / Shell scripting.
• Git / Gitlab / Github.
• Docker.
• CD/CI. Jenkins, Gitlab.
• ElasticSearch.
• Excel.

• Google Cloud. Functions, storage, big query.
• AWS.

• SQL.
• NoSQL / Relational algebra. Appears 5x less than SQL, but still interesting to learn.

Soft skills were nearly as mentionned as "Python" or "Tensorflow", so they seem really important.

• Communication. Being able to explain complex algorithms to non-technical clients or other employees. Being able to write reports and documentation on your search work.
• Self-organisation. Being able to organize your work without direct instructions.
• Business inteligence / CRM. Being able to understand how AI can improve a business and client relation management.
• Technological watch. Being able to organise and documentate a technological watch so your company and employees are always open to state of the art technics.