Decide whether to specialize in NLP, forecasting or tabular ML

A specialization is a set of problems, data constraints and evaluation habits—not a library name. Choose through small work samples and the opportunities available to you, then build depth after you have evidence about the work.

Use a score only to select the next trial

The career evidence lab combines invented interest and data-access ratings.

from career_cases import specialization_case

result = specialization_case()
assert result["recommended_trial"] == "forecasting"
assert result["recommendation_requires_project_trial"] is True
print(result["scores"])

Forecasting scores highest because the fixture combines strong interest in temporal operations with access to time-series history. NLP interest is high, but labelled-language access is weak. This is not a validated career assessment or a market-demand claim; it chooses the next experiment.

Compare the evidence each path needs

NLP: annotation rules, duplicate control, language and class slices, retrieval or extraction evaluation, source traceability and privacy. A fluent generated example is not enough.

Forecasting: forecast origins, horizon, seasonal baseline, rolling backtest, interval coverage, known-future covariates and structural-break analysis. Random splits usually answer the wrong question.

Tabular ML: row grain, label maturity, grouped or temporal splits, preprocessing boundaries, calibration, decision thresholds, error costs and drift. Strong tree models do not excuse weak problem framing.

Run matched work samples

Spend a fixed amount of time on one narrow project in each path. Use the same standards: decision brief, provenance, simple baseline, held-out design, adverse result, reproducible output and short handover. Track whether you can access meaningful data and feedback, whether you enjoy debugging its failure modes and whether target roles need that depth.

Then choose a six-to-eight-week depth project. Read foundational papers and official tools only as the project requires them. Revisit the decision after completing evidence, rather than switching specializations whenever a new model is released.

The Data Science course includes NLP, time series and tabular modelling foundations. Its draft lab results also show why specialization demands deeper evaluation than running an API or fitting one candidate.

Exercise

Create three two-hour project briefs from one domain. Score data access, decision clarity, feedback access and interest before and after each trial. Select the path whose completed evidence supports further work and document what could reverse the choice.

Continue learning

This article is part of the Data science careers and portfolio decisions sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Build a portfolio evidence index for interviewers.
• Continue with Design a capstone demo that survives difficult questions.

References: Hugging Face task documentation, statsmodels time-series analysis, and scikit-learn supervised learning.