1. Objectives
- Choose the number of clusters c with the fuzzy partition coefficient (FPC).
- Explain how the fuzzifier m makes the groups sharper or softer.
- Group the soil tests of 45 cotton fields and find the least certain fields.
- Predict the memberships of new fields with cmeans_predict and turn them into advice.
2. Word Bank
Word | Simple meaning | Real-world example |
|---|---|---|
FPC | Fuzzy Partition Coefficient: Σu²/N, from 1/c (all mixed) to 1 (hard groups) | Highest FPC -> a good c |
Fuzzifier m | Controls softness: near 1 = almost hard, larger = softer | m = 1.5, 2, 3 |
cmeans_predict | Gives memberships of NEW data using centres already found (centres are not moved) | A new soil test from the lab |
Soil pH | How acidic (< 7) or alkaline (> 7) the soil is; above about 8.5 is a problem for cotton | pH 8.6: add gypsum |
Organic matter | Rotted plant and animal material in the soil (%); more is better for water and nutrients | 0.45 % poor, 1.35 % rich |
Gypsum | Calcium sulphate; farmers add it to lower the alkalinity of soil | Punjab agriculture department advice |
3. Background
How many groups?
In P12a we were told c = 4. Usually nobody tells us. The fuzzy partition coefficient FPC = Σu²/N helps: run cmeans for several c and pick the c with the HIGHEST FPC (the crispest groups). Like the elbow and silhouette of P06a it is a guide, not a law - always check that the groups make sense to a person who knows the field.
The fuzzifier m
m decides how quickly membership falls with distance: u_i = 1 / Σ_j (d_i / d_j)^(2/(m-1)). With m close to 1 the nearest centre takes almost everything (like k-means); with m = 3 memberships are spread out. m = 2 is the usual choice.
The farm problem
The soil-testing lab at Vehari sent soil tests of 45 cotton fields near Burewala: pH and organic matter %. Fields of similar soil need similar advice - gypsum for alkaline soil, compost or green manure for poor soil. soil.csv has the 45 tests. After grouping, NEW tests can be placed in the groups with cmeans_predict, without running the clustering again.
Data files for this lab
households.csv
house,summer_units,winter_units H01,6.6,1.8 H02,5.9,3.5 H03,19.4,3.7 H04,20.7,5.9 H05,24.1,5.4 H06,8.5,4.2 H07,7.9,11.7 H08,19.6,6.1 H09,22.1,2.4 H10,4.8,2.6 ... (60 data rows in total - copy the full file from the lab folder) |
soil.csv
field,ph,organic_matter F01,8.65,0.41 F02,7.73,0.75 F03,7.8,0.84 F04,7.37,1.42 F05,7.36,1.24 F06,8.7,0.41 F07,7.79,0.85 F08,7.23,1.4 F09,7.91,0.88 F10,8.67,0.49 ... (45 data rows in total - copy the full file from the lab folder) |
4. Time Plan (50 minutes)
Minutes | Activity | Part |
|---|---|---|
0-5 | Background: choosing c with FPC; the fuzzifier m; the soil problem | 1-3 |
5-10 | Guided warm-up: FPC for c = 2 to 7; m for one house | Part A |
10-13 | Predict the output | Part B |
13-17 | Bug hunt | Part C |
17-47 | Graded tasks 1-3 | Part D |
47-50 | Wrap-up; post-lab questions set as homework | Part E |
Part A - Guided Warm-up (type, run, observe)
A1. Which c for the households?
Run cmeans on households.csv for c = 2 to 7 and print the FPC of each. Does the best c agree with P06a (k = 4)?
import numpy as np import pandas as pd import skfuzzy as fuzz df = pd.read_csv("households.csv") X = df[["summer_units", "winter_units"]].to_numpy() # 60 houses x 2 features data = X.T # skfuzzy wants features x samples: 2 x 60 for c in range(2, 8): *_, fpc = fuzz.cluster.cmeans(data, c=c, m=2, error=0.005, maxiter=1000, seed=0) print(f"c = {c}: FPC = {fpc:.3f}") |
Output c = 2: FPC = 0.856 c = 3: FPC = 0.830 c = 4: FPC = 0.878 c = 5: FPC = 0.796 c = 6: FPC = 0.733 c = 7: FPC = 0.675 |
A2. The effect of m on one house
House H23 (7.6, 6.9) was the most in-between house in P12a. Print its four memberships for m = 1.5, 2 and 3.
# ... set-up: load the households (as shown earlier) ... cntr, u, u0, d, jm, p, fpc = fuzz.cluster.cmeans(data, c=4, m=2, error=0.005, maxiter=1000, seed=0) order = np.argsort(cntr[:, 0] + cntr[:, 1]) # sort clusters from lightest to heaviest users cntr, u = cntr[order], u[order] NAMES = ["low use", "winter heater", "summer AC", "heavy all year"] for m in (1.5, 2, 3): um, *_ = fuzz.cluster.cmeans_predict(data[:, [22]], cntr, m, error=0.005, maxiter=1000, seed=0) print(f"m = {m}: " + " ".join(f"{n} {v:.2f}" for n, v in zip(NAMES, um[:, 0]))) |
Output m = 1.5: low use 0.82 winter heater 0.18 summer AC 0.01 heavy all year 0.00 m = 2: low use 0.63 winter heater 0.29 summer AC 0.05 heavy all year 0.02 m = 3: low use 0.47 winter heater 0.32 summer AC 0.13 heavy all year 0.08 |
Part B - Predict the Output
Write your prediction BEFORE running the code. Then run it and correct yourself in a different colour.
B1.
import numpy as np d = np.array([1.0, 2.0]) # distances of one item to two centres for m in (2, 3): u = [1 / np.sum((di / d) ** (2 / (m - 1))) for di in d] print(m, np.round(u, 3)) |
My prediction |
B2.
import numpy as np u = np.full((2, 6), 0.5) print(np.sum(u ** 2) / u.shape[1]) |
My prediction |
B3.
import numpy as np import pandas as pd import skfuzzy as fuzz soil = pd.read_csv("soil.csv") S = soil[["ph", "organic_matter"]].to_numpy() # 45 fields x 2 features cntr, u, u0, d, jm, p, fpc = fuzz.cluster.cmeans(S.T, c=3, m=2, error=0.005, maxiter=1000, seed=0) alk = int(np.argmax(cntr[:, 0])) # highest pH rich = int(np.argmax(cntr[:, 1])) # highest organic matter NAMES = {alk: "alkaline", rich: "compost-rich"} NAMES[3 - alk - rich] = "normal" new = np.array([[8.4, 0.50], [7.5, 1.10]]) u_new, *_ = fuzz.cluster.cmeans_predict(new.T, cntr, 2, error=0.005, maxiter=1000, seed=0) print(cntr.shape, u_new.shape) |
My prediction |
Part C - Bug Hunt
Each program has ONE bug. Some crash with an error message; others run but give a wrong answer (silent logic error - the dangerous kind). Find the bug, explain it and write the fix.
C1.
# ... set-up: load the soil tests (as shown earlier) ... fpcs = {} for c in range(2, 7): *_, fpcs[c] = fuzz.cluster.cmeans(S.T, c=c, m=2, error=0.005, maxiter=1000, seed=0) best = min(fpcs, key=fpcs.get) print("best c =", best) |
It runs and prints best c = 6 |
Bug, why it happens, and the fixed line(s) |
C2.
# ... set-up: load the soil tests (as shown earlier) ... # ... set-up: fuzzy c-means on the soil tests (c = 3) and name the groups (as shown earlier) ... new = np.array([[8.4, 0.50], [7.5, 1.10], [8.2, 0.65]]) u_new, *_ = fuzz.cluster.cmeans_predict(new, cntr, 2, error=0.005, maxiter=1000, seed=0) print(np.round(u_new, 2)) |
Bug, why it happens, and the fixed line(s) |
C3.
# ... set-up: load the soil tests (as shown earlier) ... # ... set-up: fuzzy c-means on the soil tests (c = 3) and name the groups (as shown earlier) ... new = np.array([[8.2, 0.65]]) u_new, *_ = fuzz.cluster.cmeans_predict(new.T, cntr, 3, error=0.005, maxiter=1000, seed=0) print({NAMES[k]: round(float(u_new[k, 0]), 2) for k in range(3)}) |
It runs and prints {'compost-rich': 0.17, 'alkaline': 0.39, 'normal': 0.44} |
Bug, why it happens, and the fixed line(s) |
Part D - Graded Tasks (10 marks)
Task | Level | Title | Marks |
|---|---|---|---|
T1 | Easy | How many kinds of field? | 3 |
T2 | Medium | Name the groups and find the least certain fields | 3 |
T3 | Challenge | Advice for three new fields | 4 |
Total | 10 |
Task 1 (Easy, 3 marks): How many kinds of field?
Load soil.csv. Run fuzzy c-means (m = 2, error = 0.005, maxiter = 1000, seed = 0) for c = 2 to 6, print the FPC of each, choose the best c and plot FPC against c (task1.png).
Sample run (your output must match):
Output (45, 3) field ph organic_matter F01 8.65 0.41 F02 7.73 0.75 F03 7.80 0.84 c = 2: FPC = 0.864 c = 3: FPC = 0.934 c = 4: FPC = 0.828 c = 5: FPC = 0.814 c = 6: FPC = 0.790 best c = 3 Saved task1.png |
Your code (or attach the file) |
Task 2 (Medium, 3 marks): Name the groups and find the least certain fields
Cluster the fields with c = 3, m = 2. Name the clusters from their centres (highest pH = alkaline, highest organic matter = compost-rich, the other = normal) and print each centre. Harden the memberships and print how many fields are in each group. Then list the three LEAST certain fields (smallest largest membership) with their two strongest groups.
Sample run (your output must match):
Output compost-rich centre: pH 7.30, organic matter 1.32 % alkaline centre: pH 8.64, organic matter 0.46 % normal centre: pH 7.80, organic matter 0.80 % fields per group: {'compost-rich': 15, 'alkaline': 15, 'normal': 15} F32 pH 7.57, OM 0.81: normal 0.83, compost-rich 0.13 F22 pH 7.99, OM 0.70: normal 0.87, alkaline 0.09 F28 pH 7.69, OM 0.94: normal 0.89, compost-rich 0.09 |
Your code (or attach the file) |
Task 3 (Challenge, 4 marks): Advice for three new fields
The lab sends three new tests: A (pH 8.4, OM 0.50), B (7.5, 1.10), C (8.2, 0.65). Use cmeans_predict with the centres from Task 2 and m = 2 to get their memberships. Print advice for each: 'add gypsum' if alkaline membership > 0.3, 'add compost / green manure' if normal + alkaline membership > 0.6 (poor organic matter), otherwise 'keep current practice'. Plot the 45 fields coloured by group and the new fields as black triangles labelled A, B, C (task3.png).
Sample run (your output must match):
Output A pH 8.4, OM 0.5: compost-rich 0.03, alkaline 0.86, normal 0.11 -> add gypsum + add compost / green manure B pH 7.5, OM 1.1: compost-rich 0.65, alkaline 0.03, normal 0.32 -> keep current practice C pH 8.2, OM 0.65: compost-rich 0.08, alkaline 0.41, normal 0.51 -> add gypsum + add compost / green manure Saved task3.png |
Your code (or attach the file) |
Part E - Post-Lab Questions (finish as homework if time runs out)
E1. How do you choose c with the FPC, and why is it only a guide?
Your answer |
E2. What happens to the memberships when m is increased from 1.5 to 3?
Your answer |
E3. What is the difference between cmeans and cmeans_predict?
Your answer |
Marking Rubric (applied to Part D)
Criterion | Full | Partial | None |
|---|---|---|---|
Code runs without errors | Runs fully | Small fixable error | Does not run |
FPC for several c, best c and plot | All correct | One slip | Several wrong |
Groups named from centres, sizes and mixed fields | Correct and explained | Partly correct | Missing |
Predictions for new fields, advice and plot | Clear and correct | Partly correct | Missing |
Submission
- Make a folder RollNo_P12b with task1.py, task2.py, task3.py (or P12b.ipynb), task1.png and task3.png.
- Upload the zipped folder to the LMS / Google Classroom by the end of the day.
- Copy Task 3 code and output into your Lab Book (Practical 12b page).
- P13 (next week) saves a fuzzy system to a file and runs it as a program - keep your fuzzy code tidy.