Classification vs Regression
Compare logistic classification with linear regression on crypto returns, and explore how win rate, return magnitude, and trading performance relate.
Run Python, change assumptions, inspect results, and learn how research becomes into a profitable trading strategy
returns = df['close'].pct_change()
signal = model.predict(features)
p_value = test(edge).pvalue
# test, challenge, refine
Choose how you learn
Own a focused research notebook for life, or subscribe for the complete MemLabs curriculum, every premium notebook, competitions, exercises, and new releases.
Buy individually
£50
per premium notebook
One payment with lifetime access to the notebook you choose.
Choose a notebookMemLabs All Access
£200/month
Every notebook and course
Access the entire platform and everything released while you are subscribed.
Notebook library
Every notebook is editable, executable, and designed around a real quantitative question.
Compare logistic classification with linear regression on crypto returns, and explore how win rate, return magnitude, and trading performance relate.
The live Hyperliquid bot behind the conditional mean strategy: how a basic statistical forecast becomes code that trades.
Build the kind of mean reversion model used at quantitative firms using basic statistics and mathematics instead of machine learning.
An example of researching a basic statistical anomaly using high-school statistics.
Explore different models that can adapt to changes in the distribution of price returns (regime changes).
Find a predictive edge in other assets' lagged returns when an asset's own past returns have none.
Model return direction and magnitude separately, compare decomposed forecasts with direct regression, and use the predictions to set market-making quotes.
Can stacking models on top of each other improve model performance?
Understand prediction error, compare MAE and MSE, and explore how large misses affect model scores using ALGO daily return forecasts.
Train an XGBoost model on calendar features, validate its predictions, and test whether the apparent edge remains after fees.
Condition the expected return on a basic statistic, without machine learning, and test whether it beats the unconditional mean benchmark.
Can the average daily return of a coin be traded, or is it only a benchmark to beat?
The live Hyperliquid bot behind the unconditional mean strategy: how a research idea becomes code that trades.
Learn how an in-sample relationship can fail to generalise by comparing Bitcoin returns across training and testing periods.
Launch a configured Python notebook directly in your browser.
Change the code, assumptions, features, and execution model.
Inspect the evidence and decide whether the result is robust enough to matter.