lev-os Greedy Algorithms Paradigm
What It Is
A problem-solving strategy that makes the locally optimal choice at each step with the hope of finding a global optimum. Unlike dynamic programming which considers all possible decisions, greedy algorithms commit to the best immediate option without reconsidering past choices. The key insight: for certain problems, always choosing what looks best right now leads to the overall best solution.
When to Use It
- Problem has greedy choice property (local optimal choices lead to global optimum)
- Problem has optimal substructure (optimal solution contains optimal subsolutions)
- Need fast, simple solution (greedy is usually O(n log n) or better)
- Can prove that greedy choice never needs to be undone
- Making irreversible decisions is acceptable (no backtracking needed)
- Problem involves optimization with ordering/scheduling/selection
Key Test: Can you prove that choosing the best option now won't block a better overall solution later?
Execution Steps
1. Evaluate Available Options
At each decision point, identify all valid choices. Define what "best" means for your problem (largest, smallest, earliest deadline, highest ratio). Sort or prioritize options if needed.
Action: List all candidates. Define the selection criterion clearly (e.g., "choose activity with earliest finish time" or "select item with highest value/weight ratio").
2. Make the Greedy Choice
Select the option that appears optimal according to your criterion. Commit to this choice immediately without considering future consequences. This is the core greedy step.
Action: Pick the best option from available candidates. Add it to your solution. This choice is final and irreversible.
3. Update State
Remove the chosen option from consideration. Update the problem state based on your choice. Define the new subproblem that remains.
Action: Mark chosen item as used. Update constraints (remaining capacity, time, etc.). Recompute what's now available for next iteration.
4. Iterate
Repeat steps 1-3 until reaching a solution or no further progress is possible. Each iteration solves a smaller subproblem.
Action: Loop until goal achieved, resources exhausted, or no valid choices remain. Return accumulated solution.
5. Verify Optimality
Prove (or test) that your greedy strategy produces optimal results. Common proof techniques: exchange argument, greedy stays ahead, structural induction.
Action: For known problems (MST, shortest path), use established proofs. For new problems, construct counterexamples to test, or prove correctness formally.
Real-World Applications
Scheduling & Resource Allocation
- Activity selection: conference room booking, CPU task scheduling
- Job sequencing: minimize weighted completion time, maximize throughput
- Huffman coding: optimal prefix-free codes for data compression
- Interval scheduling: maximize non-overlapping meetings
Graph Algorithms
- Dijkstra's shortest path: GPS navigation, network routing (OSPF protocol)
- Prim's/Kruskal's MST: network design, circuit layout, cluster analysis
- Fractional knapsack: portfolio optimization with divisible assets
Network & Routing
- Load balancing: distribute requests to servers (least loaded server)
- Bandwidth allocation: maximize network utilization
- Packet routing: next-hop forwarding (shortest path to destination)
Financial & Trading
- Coin change with specific denominations: make change with fewest coins
- Stock trading: maximize profit with one buy-sell (best time to buy/sell)
- Fractional resource allocation: allocate budget to projects by ROI
Data Compression
- Huffman trees: ZIP, JPEG, MP3 compression
- Run-length encoding: simple compression for repeated data
- Optimal merge patterns: external sorting of sorted runs
Machine Learning
- Feature selection: forward/backward selection by information gain
- K-means clustering: assign points to nearest centroid
- Decision tree splitting: choose best split by Gini/entropy
Anti-Patterns
Using greedy when it doesn't guarantee optimality → Produces suboptimal results; verify greedy choice property first or use DP.
Not sorting when needed → Greedy algorithms often require sorted input; missing this step breaks the strategy.
Confusing greedy with dynamic programming → If choices overlap or need reconsideration, use DP; greedy is for irreversible local decisions.
Applying to 0/1 knapsack → Greedy fails here (DP required); only works for fractional knapsack.
Ignoring problem constraints → Greedy choice must respect all constraints; check feasibility before committing.
No proof of correctness → Just because greedy seems to work doesn't mean it's optimal; always verify with proof or exhaustive testing.
Success Metrics
- Solution matches known optimal value (for problems with established answers)
- Algorithm runs in expected time (typically O(n log n) for sorting-based greedy)
- Proof of correctness established (exchange argument, greedy stays ahead)
- Empirical testing on edge cases confirms optimality
- Simplicity: code is straightforward, easy to understand and maintain
Related Frameworks
- Dynamic Programming: When greedy fails, DP often succeeds (but slower)
- Divide and Conquer: Both make decisions and recurse, but D&C doesn't commit locally
- Backtracking: For constraint satisfaction when greedy is insufficient
- Primal-Dual Algorithms: Greedy with duality theory for approximation algorithms
- Matroid Theory: Mathematical framework proving when greedy works
Common Pitfalls
- Assuming greedy always works (it doesn't - most optimization problems need DP or search)
- Not proving correctness (greedy algorithms are easy to get wrong)
- Choosing wrong greedy criterion (e.g., sorting by value instead of value/weight ratio)
- Forgetting to sort input when order matters
- Applying greedy to problems requiring backtracking (e.g., N-Queens, Sudoku)
- Not handling ties in selection criterion (can lead to non-deterministic results)
Tools & Resources
- Visualization: VisuAlgo (visualgo.net) for greedy algorithm animations
- Practice: LeetCode greedy tag (~150 problems), Codeforces greedy category
- Books: "Algorithm Design" (Kleinberg & Tardos), "Introduction to Algorithms" (CLRS)
- Proof Techniques: "Exchange argument" tutorial, "Greedy stays ahead" proof pattern
- Counterexample Testing: Generate random inputs, compare greedy vs. brute force on small instances
Classic Greedy Algorithms
Activity Selection: Choose activities with earliest finish time → maximizes number of non-overlapping activities
Fractional Knapsack: Sort items by value/weight ratio, take highest ratios first → optimal value
Huffman Coding: Build tree by repeatedly merging two lowest-frequency nodes → optimal prefix-free code
Dijkstra's Algorithm: Always expand nearest unvisited node → shortest paths from source
Kruskal's MST: Add cheapest edge that doesn't create cycle → minimum spanning tree
Prim's MST: Grow tree by adding cheapest edge to current tree → minimum spanning tree
Coin Change (specific denominations): Always take largest coin ≤ remaining amount → fewest coins (works for standard denominations like US coins)
When Greedy Fails
0/1 Knapsack: Can't split items, so greedy by value/weight fails. Need DP.
Longest Path: Greedy (always choose longest available edge) fails. NP-hard, need exponential search.
Traveling Salesman: Greedy (nearest neighbor) gives 2-approximation but not optimal. Need sophisticated approaches.
Job Scheduling with Dependencies: Greedy ignores dependencies. Need topological sort + DP.
Fair Division: Greedy may give one party too much. Need game-theoretic approaches.
Framework Type: Algorithm Design Paradigm Domain: Computer Science, Optimization Practitioner Score: 9/10 - Powers many production algorithms (Dijkstra, MST, Huffman) Complexity: Medium - Conceptually simple, but proving correctness is non-trivial Prerequisites: Sorting algorithms, basic graph theory, proof techniques (induction, contradiction)