Deep Signal Lab Interactive Explorer
This explorer lets you play with two of the core Deep Signal Lab views in the browser: a time series with anomaly flags and the tail of the power distribution. Move the threshold and watch how the alerts and histogram change, mirroring the ideas from the full notebook analysis.
Time series and anomaly view
5 / 220 points flaggedBlue line shows the simulated signal. Pink dots are points whose value is above the chosen z score threshold relative to the overall distribution of this series.
Threshold control
Lower thresholds flag more points (higher sensitivity but more false positives). Higher thresholds flag only the strongest spikes (more conservative alerts).
Interpretation
This is a simplified version of the anomaly detector from the notebook: we compute a global mean and standard deviation for the series, then flag points whose value is more than a chosen number of standard deviations above that mean.
In the full analysis, we use a more robust per-frequency approach (median + MAD) on a time–frequency grid; this interactive view gives recruiters an intuitive sense of how tuning the threshold changes what gets flagged.
Threshold trade off curve
Shows how many alerts you get at different thresholds. The pink dot marks your current setting. Lower thresholds = more alerts (higher sensitivity, more false positives).
Mini spectrogram (time × frequency)
A simplified 2D view of signal power over time and frequency bands. Bright regions with accent color are flagged as anomalies at the current threshold. This mirrors the spectrogram style from the full analysis.
Quick stats
What does all of this mean?
The core problem
In real world data analysis, you often need to monitor a stream of measurements and flag moments when something unusual happens. This could be a spike in server errors, a sudden change in user behavior, or in this case, a potential signal of interest in noisy radio telescope data. The challenge is deciding what counts as "unusual" without getting overwhelmed by false alarms.
What you're seeing here
- Time series view: Shows the signal over time. The blue line is the actual measurement, and pink dots mark points that exceed your chosen threshold. This is the "monitoring dashboard" view, you can see at a glance when something interesting happened.
- Distribution view: Shows how often different power levels occur. Most values cluster in the middle (normal noise), but a few sit in the extreme right tail (potential anomalies). Bars tinted with the accent color contain flagged points. This view helps you understand where your threshold sits relative to the bulk of the data.
- Threshold trade off curve: This is the key insight for any anomaly detection system. It shows that as you lower the threshold, you catch more potential signals but also flag more false positives. As you raise it, you become more conservative but might miss real events. The pink dot shows where you are on this curve right now.
- Mini spectrogram: A 2D view showing signal power across both time and frequency bands. This is how radio astronomers actually visualize their data. Bright regions with accent color are flagged as anomalies. This connects the interactive explorer to the full Deep Signal Lab analysis, where we work with time frequency grids.
- Quick stats: Real-time metrics that update as you adjust the threshold. The percentage flagged tells you how aggressive your detector is, and the max z score shows how extreme the most unusual point in the series is.
Why this matters for data analysis
This interactive explorer demonstrates a fundamental skill in data analysis: understanding the trade offs in your detection rules. In production systems, you need to balance sensitivity (catching real events) with specificity (avoiding false alarms). The threshold trade off curve makes this explicit, you can see exactly how many alerts you'll get at different sensitivity levels. This same pattern appears in fraud detection, system monitoring, quality control, and many other domains where you're looking for needles in haystacks.
How this connects to the full analysis
The full Deep Signal Lab notebook uses a more sophisticated approach: instead of a single global threshold, we compute robust z scores per frequency band (using median and MAD instead of mean and standard deviation). This handles cases where different frequency bands have different baseline noise levels. But the core idea is the same: flag points that are unusually strong relative to their local context. This interactive explorer gives you an intuitive feel for how threshold tuning works, which is the foundation for understanding the more complex methods in the notebook.