Online notebook execution#
The Cancer Live-Coding Notebook offers Classification (default) and Survival Analysis in the Choose a notebook dropdown. Each selection opens its own documentation page with the selected notebook’s cells, contents and launch links. Both can be read as a static page, connected to a temporary Binder kernel through Thebe, or opened in Google Colab. No code is run by opening the documentation page.
Run on this website#
Choose Run interactively on this website and wait for the connection status to show that the kernel is connected. The first Binder build may take several minutes. Code cells become editable, and you can run them individually, in order. A connection alone does not execute the notebook or verify that every research dependency works.
The kernel is configured to start in notebooks/MDAS/cancer/, so the companion synthetic_data_generation.py module can be imported. Both notebooks’ ../../../data/synthetic_data path resolves to data/synthetic_data inside the temporary repository checkout. The helper generates synthetic data there; no patient dataset is needed.
Binder uses the repository on GitHub, not your local checkout. The organiser has confirmed a successful remote launch and Thebe kernel connection after publishing the Binder configuration. That does not establish full research-training validation or validate Colab.
Switch notebooks and save edits#
Switching loads a new page and starts with its saved cells; kernels and variables are not transferred. After interactive launch, switching asks for confirmation. Cancel to stay on the current page and choose Download notebook with edits before leaving. The download includes current code edits and original Markdown, with outputs cleared; it does not include kernel variables or generated files. Check that the download succeeded. Copy important results or download generated files separately. Reloading, closing the tab and following documentation links also trigger the browser’s leave-page warning after launch.
Each notebook links its corresponding _blank.ipynb as an Optional blank exercise on GitHub. These exercise notebooks are not the displayed defaults.
Connection troubleshooting#
If the launch reports No environment specification found, Binder has not reached package installation or kernel startup. It cannot see a usable environment definition in the selected public revision. Installing more packages locally does not fix that remote failure.
The launch control now checks the public revision for .binder/runtime.txt, .binder/requirements.txt, the notebook and its companion module before requesting Binder. If files are absent, the status names them and explains that local files are not sent to Binder. Once present, it launches the exact commit checked. A GitHub API timeout, blocked request or rate limit is reported separately and is not treated as evidence that packages failed to install.
Open the Binder build logs link on the notebook page to distinguish repository retrieval, environment discovery/build, server launch and kernel connection. A generic Thebe failed status alone does not identify the stage. A successful environment-file check is not a guarantee that the image or kernel will work.
The earlier revision ed163a322186055c8c5b79885288ad9a051f89c4 lacked Binder files. The organiser subsequently published them in 090f1ff68a57cd2f04ab1df9008fd802e6caa278 and confirmed a ready server and working Thebe connection. See the maintainer build guide for diagnostics; a future failure still needs its actual launch stage checked.
Open in Google Colab#
Use the Open in Google Colab button on the selected notebook page: Classification or Survival Analysis.
Each link opens its original MDAS Cancer notebook. It does not automatically install dependencies or retrieve the companion synthetic_data_generation.py module. A Colab-ready environment is deferred for follow-up work; use Run interactively on this website for the currently working Binder option.
Compute and privacy limits#
The helper defaults to 100 synthetic subjects and 20–199 patches per subject, each with 32 features. That feature data is small, but importing the ML stack, training for 40 epochs, fitting the optional baseline and computing bootstrap intervals need additional memory and CPU time. The entire notebook has not been benchmarked on public Binder. Run training deliberately; avoid repeated or simultaneous advanced runs.
Binder resources are limited and sessions are temporary; availability and completion are not guaranteed. Download work you need to keep: edits and generated files in temporary online sessions may be lost.
Caution
Never upload identifiable patient data, confidential research data, private datasets, credentials or other secrets to public execution services. Synthetic workflow scores are not clinical evidence.