CITC University Course Timetabling with SDCSA

Thesis Project — University of Science and Technology of the Philippines (USTP) Department: College of Information Technology and Computing (CITC) Algorithm: (SUBSTRATE DRIFT CLONAL SELECTION ALGORITHM) SDCSA Validation Standard: ITC 2019-inspired constraint framework

---

Pipeline Overview

CITC SCHED (original)                     Structured_Data (teammate's extract)
        |                                           |
        +-------------------------------------------+
                            |
                      preprocess.py
                            |
                     cleaned/ folder
                  (normalized data)
                            |
            cleaned/Algorithm_Input_Cleaned.xlsx
                            |
                   Upload to app.py (Streamlit)
                            |
                    ETFCSA-TSD Optimizer
                            |
                  +--------------------+
                  |                    |
          Optimized_Schedule.xlsx   ITC 2019 XML
                  |
          validate_schedule.py
          (post-optimization check)
                  |
              logs/ folder
          (run_YYYYMMDD_HHMMSS.log)

---

File Structure

File/Folder	Purpose
CITC SCHED/	Original department schedule (100 section sheets). Do not modify.
Structured_Data_Algorithm_Input_Refined.xlsx	Teammate's original input file. Do not modify.
preprocess.py	Cleans and normalizes both source files into cleaned/
cleaned/	Output of preprocessing — normalized, ready-to-use data
cleaned/CITC_SCHEDULING_CLEANED.xlsx	Cleaned schedule (100 sheets, all 755 meetings)
cleaned/Algorithm_Input_Cleaned.xlsx	Optimizer input (Class_Requirements + Room_List + Section_List + Instructors)
app.py	Streamlit web app — runs ETFCSA-TSD optimization with ITC 2019 constraints
tsd.py	ETFCSA-TSD algorithm implementation (generic optimizer, not modified)
validate_schedule.py	ITC 2019-style validator — checks hard/soft constraints independently
logs/	Auto-generated optimization run logs (one per run)
preprocessing_documentation.md	Documents all data quality issues and normalization decisions
validation_documentation.md	Documents ITC 2019 constraint definitions, online class handling, initial results
optimization_documentation.md	Documents encoding, objective function, parameter tuning, logging

---

Step-by-Step Usage

1. Preprocessing

python preprocess.py

What it does:

• Reads the original CITC SCHED/ Excel (100 section sheets) and Structured_Data_Algorithm_Input_Refined.xlsx
• Normalizes instructor names (29 duplicates), subject codes (5), room names (8)
• Detects online classes (Room is NaN = online, 202 of 755 meetings)
• Preserves all 755 multi-day meetings (original input had collapsed them to 488)
• Outputs cleaned files to cleaned/ folder

Details: See preprocessing_documentation.md

2. Validation of Initial Schedule

app_visualized_labeled.py

What it does:

• Validates the original CITC department schedule against ITC 2019-style constraints
• Establishes a baseline: 380 hard violations (223 room + 144 instructor + 13 section), 245 soft violations, INFEASIBLE

Details: See validation_documentation.md

3. Optimization

streamlit run app.py

Steps in the app:

1. Upload cleaned/Algorithm_Input_Cleaned.xlsx
2. Adjust algorithm parameters in sidebar (population, max evals, clone rate, rho)
3. Adjust soft constraint weights (or set all to 0 for feasibility-first runs)
4. Click Run Optimization
5. Section 3 — Validation: Review before-vs-after comparison, hard constraint breakdown (H1/H2/H3 in columns), soft constraint breakdown (S1-S5 with DISABLED indicators if weight=0)
6. Section 4 — Download: Optimized_Schedule.xlsx, ITC 2019 XML, full validation report
7. A run log is auto-saved to logs/

Details: See optimization_documentation.md

4. Review Logs

After each optimization run, a log file is saved to logs/run_YYYYMMDD_HHMMSS.log containing:

• Runtime, final fitness
• Algorithm parameters and soft weights used
• Problem size (classes, variables, rooms, timeslots)
• Validation summary (feasibility, hard/soft breakdown by constraint)

Compare logs across runs to identify the best parameter configuration.

---

Constraint Summary

Hard Constraints (must be 0 for feasibility)

ID	Constraint	Applies To
H1	Room Conflict — no two f2f classes in the same room at the same time	Face-to-face only
H2	Instructor Conflict — no instructor teaches two classes at the same time	All classes
H3	Section Conflict — no section attends two classes at the same time	All classes

Soft Constraints (minimize for quality)

ID	Constraint	Default Weight
S1	Room Type Mismatch (lab subject in non-lab room)	0.5
S2	Section Daily Overload (> 5 classes/day)	0.3
S3	Instructor Daily Overload (> 5 classes/day)	0.3
S4	Schedule Gap (> 2h between consecutive classes)	0.2
S5	Late Class (starts at or after 6:00 PM)	0.1

---

Parameter Tuning Guide

Parameter	Range	Notes
Population (N)	200-300	Higher = more diversity, slower per generation
Max Evaluations	150,000-500,000	More evals = better solutions, longer runtime
Clones (n_clones)	10-15	More clones = more exploitation of good solutions
Rho (decay)	0.95-0.98	Lower = faster forgetting, helps escape local optima

Recommended strategy:

1. Phase 1 (Feasibility): Set all soft weights to 0. Focus on eliminating hard violations.
2. Phase 2 (Quality): Once hard violations = 0, re-enable soft weights and run again.

---

Dependencies

pip install streamlit pandas numpy openpyxl xlsxwriter requests

---

Data Summary

Metric	Value
Total class meetings	755
Face-to-face meetings	553 (73%)
Online meetings	202 (27%)
Unique sections	100
Unique instructors	170 (after normalization)
Unique subjects	82 (after normalization)
Unique rooms	69 (after normalization)
Decision variables	1,308 (553x2 + 202x1)
Timeslots	60 (6 days x 10 periods)