This Grid Search lesson is for a beginner. Read the short notes, run each example, then change one name or number and run it again.
Grid search tries a list of settings and keeps the one with the best score on held-out rows.
The example scores two thresholds and keeps the higher accuracy.
Example 1: Start here
Score each setting on the validation rows, not on the final test rows.
actual = [1, 0, 1, 0] # a list; the values stay in this order
probs = [0.9, 0.2, 0.4, 0.7] # a list; the values stay in this order
best = None # no value yet
for threshold in (0.3, 0.6): # repeat the indented lines once per value
pred = [1 if p >= threshold else 0 for p in probs] # a list; the values stay in this order
hits = sum(a == p for a, p in zip(actual, pred)) # the total
if best is None or hits > best[0]: # run the indented lines only when this is true
best = (hits, threshold) # a tuple; a fixed group of values
print(best)
Example 2: Add two rows
Grid Search applies one operation to every number. zip pairs the rows, and the list keeps each sum.
left = [1, 2, 20] # a list; the values stay in this order
right = [10, 20, 30] # a list; the values stay in this order
print([a + b for a, b in zip(left, right)]) # pair the lists, then combine each pair
Example 3: Average a column
The average is the total divided by how many numbers you have. That is the same summary a statistics page starts with.
scores = [70, 20, 90] # a list; the values stay in this order
print(sum(scores) / len(scores)) # len counts the items
Example 4: Keep values above a line
A cutoff drops the low values. Change 80 and run the example again to see who remains.
scores = [70, 20, 90] # a list; the values stay in this order
print([score for score in scores if score >= 80]) # keep the items that pass the test
Score each setting on the validation rows, not on the final test rows.
