Skip to content

RUN_STREAMLIT_DEMO.md

Latest commit

 

History

History
176 lines (127 loc) · 5.65 KB

RUN_STREAMLIT_DEMO.md

File metadata and controls

176 lines (127 loc) · 5.65 KB

HIFUN 4-Min Video Script (Fast Terminal + Strict UI)

Use this guide to record a clean 4-minute project demo with:

  • Fast mode findings shown in terminal.
  • Strict mode findings shown in Streamlit UI.

All commands are from project root.

1) Pre-record Setup (2-3 minutes before recording)

cd /home/mitta/HIFUN-Router-clone
source .venv/bin/activate
export PYTHONPATH="$PWD"

Install Streamlit once if needed:

pip install streamlit==1.44.1

2) Prepare fast artifacts for terminal section

This generates quick runtime artifacts used by the dashboard fast profile.

# Quick shift artifact
python experiments/dataset_shift_evaluation.py \
	--source training_data/real_labeled_runs_strict_curated.csv \
	--out_json experiments/results/dataset_shift_eval_fast_runtime.json \
	--out_md experiments/results/dataset_shift_eval_fast_runtime.md

# Quick robustness artifact (reduced bootstrap/permutation counts)
python experiments/strict_robustness_evaluation.py \
	--train training_data/fixed_train_base_strict.csv \
	--eval training_data/fixed_eval_set_strict.csv \
	--transfer_source training_data/real_labeled_runs_strict_curated.csv \
	--n_bootstrap 150 \
	--n_perm_labels 20 \
	--n_perm_features 8 \
	--out_json experiments/results/strict_robustness_eval_fast_runtime.json \
	--out_md experiments/results/strict_robustness_eval_fast_runtime.md

Optional strict refresh (longer, skip during recording):

./run_project_strict.sh smoke

3) Start dashboard

streamlit run streamlit_app.py --server.port 8501 --server.headless true

Open:

http://localhost:8501

4) Recording script (4:00 total)

Segment A (0:00-0:20) - Intro

On screen:

  • Show project root in terminal and Streamlit tab open.

Say: "This is HIFUN Router, a hybrid SQL-vs-GRAPH routing system. In this demo, I show quick terminal findings in fast mode, then strict evidence in the dashboard for final reporting quality."

Segment B (0:20-1:25) - Fast mode findings in terminal

On screen:

  • Terminal only.
  • Run and show command outputs with saved artifact paths.

Commands to run live:

python experiments/dataset_shift_evaluation.py \
	--source training_data/real_labeled_runs_strict_curated.csv \
	--out_json experiments/results/dataset_shift_eval_fast_runtime.json \
	--out_md experiments/results/dataset_shift_eval_fast_runtime.md

python experiments/strict_robustness_evaluation.py \
	--train training_data/fixed_train_base_strict.csv \
	--eval training_data/fixed_eval_set_strict.csv \
	--transfer_source training_data/real_labeled_runs_strict_curated.csv \
	--n_bootstrap 150 \
	--n_perm_labels 20 \
	--n_perm_features 8 \
	--out_json experiments/results/strict_robustness_eval_fast_runtime.json \
	--out_md experiments/results/strict_robustness_eval_fast_runtime.md

Say: "Fast mode is for quick iteration. We reduce robustness sampling so we can quickly validate trend direction and artifact health before strict reporting."

Optional one-line verification:

ls -lh experiments/results/*fast_runtime.json experiments/results/*fast_runtime.md

Segment C (1:25-1:40) - Transition from terminal to UI

On screen:

  • Switch to browser with Streamlit app.
  • Sidebar: select fast briefly to show that quick artifacts are loaded.

Say: "The dashboard can load this fast profile for quick checks. For publication-grade findings, we switch to strict profile."

Segment D (1:40-3:35) - Strict mode findings in UI

On screen:

  • Sidebar: select strict.
  • Keep Compact view enabled.
  • Walk tabs in this order:
    1. Dataset and Quality
    2. Relevance Evaluation
    3. Robustness Evaluation
    4. Cross-Dataset Generalization

Say (suggested lines):

  1. Dataset and Quality (1:40-2:00) "This summarizes train/eval size, label distribution, and quality gates. We use strict curated real measurements for reliable claims."

  2. Relevance Evaluation (2:00-2:35) "Here are model-level routing metrics. The key finding is parity between learned routing and the traversal rule under strict evaluation, which supports robustness of the decision boundary."

  3. Robustness Evaluation (2:35-3:05) "Bootstrap confidence intervals and permutation sanity checks indicate the model signal is stable and not due to leakage."

  4. Cross-Dataset Generalization (3:05-3:35) "Cross-dataset transfer shows where routing generalizes and where domain shift remains challenging."

Segment E (3:35-4:00) - Conclusion

On screen:

  • Return to Executive View.

Say: "In summary, fast mode supports rapid iteration in terminal, and strict mode provides trustworthy dashboard evidence for reporting. Next steps are improving graph-class coverage and stronger cross-domain transfer."

5) Dashboard utilization checklist (during recording)

  • Use sidebar Run Profile:
    • fast for quick runtime artifacts.
    • strict for final reporting artifacts.
  • Keep Compact view ON during recording to avoid extra detail.
  • In Artifacts sidebar section, briefly show loaded file paths to reinforce reproducibility.
  • Prefer stable tab order: Dataset and Quality -> Relevance Evaluation -> Robustness Evaluation -> Cross-Dataset Generalization.
  • Avoid scrolling too much; pause 2-3 seconds per key metric card.

6) Camera and pacing tips

  • Record at 1080p, 16:9, terminal zoom >= 125%.
  • Use one terminal pane and one browser tab only.
  • Keep command copy/paste prepared in a scratchpad to avoid typing delays.
  • Speak at 130-150 words/minute so the 4-minute target is maintained.

7) Stop dashboard

Press Ctrl+C in the Streamlit terminal.

Optional: Run dashboard without activating shell

cd /home/mitta/HIFUN-Router-clone
.venv/bin/streamlit run streamlit_app.py --server.port 8501 --server.headless true