Tools and Methods: learner worksheet

Lecture plan and slides · Teaching sequence

35 minutes plus peer review. Work in pairs. Every value and identifier here is invented teaching data, not MICrONS, H01 or a real materialization version. The paper route requires only a calculator. The optional code route needs Python 3, with no additional packages, network access or credentials.

1. Reproduce a count (15 minutes)

Count rows with post = 101, region = core, and score ≥ 0.80. Treat the score as a toy filter, not a calibrated probability. Each ID appears once per snapshot.

Snapshot A contains these five rows, written as (id, post, region, score):

(s1, 101, core,    0.95)
(s2, 101, core,    0.80)
(s3, 101, core,    0.60)
(s4, 101, outside, 0.99)
(s5, 102, core,    0.90)

Snapshot B retains all five IDs, changes s3’s score to 0.90, and adds (s6, 101, core, 0.85). For this exercise the target and region labels have the same meaning in both snapshots. Real reconstruction IDs require version-aware checking.

List the included IDs and excluded rows for each snapshot. Explain each difference. Would changing ≥ to > affect the result? Can the change establish biological growth?

Optional runnable route: save the teaching query as tools-and-methods-query.py in a working folder, then run:

python3 tools-and-methods-query.py --version snapshot-a
python3 tools-and-methods-query.py --version snapshot-b

The program embeds the exact rows printed above and reports the filters, IDs, counts, Python version and a SHA-256 hash of its source. It refuses an unspecified version or latest. Save each output with the exact command and execution date. This is an offline teaching fixture, not a public-volume query or a CAVE client.

2. Repair the methods record (10 minutes)

A colleague’s methods say: “We downloaded the latest synapses for cell 101 and counted the high-confidence inputs.” Rewrite this as a reproducible record for Snapshot A. Include dataset/snapshot, object and region scope, exact filter, query code identity, execution environment/date, returned IDs and one limitation.

For a real dataset, what would you additionally record about segmentation state, coordinate units/frame, table schema and access? What could you check if an earlier run’s version was missing? Distinguish evidence you can recover from what you would have to label unknown.

3. Estimate capacity (10 minutes)

For a separate hypothetical acquisition, assume a 100 × 100 × 40 µm volume, 10 × 10 × 40 nm voxels, one channel, 8-bit unsigned samples and no compression.

Calculate the voxel dimensions, total voxels, raw bytes and decimal GB. Estimate capacity for three total copies of that raw image. If sustained payload transfer were 100 MB/s, what is the ideal transfer time for one copy? Use 1 GB = 10⁹ bytes and 1 MB = 10⁶ bytes. Name at least three things omitted from this estimate.

Peer review and exit ticket

Exchange methods records and reproduce the selected IDs without asking the author what they meant. Circle any missing decision. Record the result of this re-run, including mismatches rather than silently correcting them.

Submit the two counts, repaired methods record and capacity estimate. Finish: “If this count changes next month, the first evidence I would compare is ___.”

Instructor model responses are public; attempt the exercise first. Teaching material: CC BY-SA 4.0, NeuroTrailblazers.