A-eye¶
You're building an image search engine called A-eye 
. You're using Mean Average Precision at K (MAP@K) to
evaluate the performance of your results.
Your dataset contains three queries and their predicted relevance on 1000 images. Calculate MAP@5 for this data.
rng = np.random.default_rng(2703)
df = pd.DataFrame({
'id': np.tile(range(1000), reps=3),
'query': np.repeat(["dolphin", "emu", "gorilla"], repeats=1000),
'pred_score': rng.uniform(low=0, high=1, size=3000),
})
df["relevant"] = False
df.loc[(df.id.isin(rng.choice(1000, size=25, replace=False))) & (df["query"] == "dolphin"), "relevant"] = True
df.loc[(df.id.isin(rng.choice(1000, size=25, replace=False))) & (df["query"] == "emu"), "relevant"] = True
df.loc[(df.id.isin(rng.choice(1000, size=25, replace=False))) & (df["query"] == "gorilla"), "relevant"] = True
print(df)
# id query pred_score relevant
# 0 0 dolphin 0.690034 False
# 1 1 dolphin 0.555039 False
# 2 2 dolphin 0.001131 False
# 3 3 dolphin 0.453131 False
# 4 4 dolphin 0.942450 False
# ... ... ... ...
# 2995 995 gorilla 0.650423 False
# 2996 996 gorilla 0.883602 False
# 2997 997 gorilla 0.756802 False
# 2998 998 gorilla 0.143045 False
# 2999 999 gorilla 0.944064 False
library(data.table)
set.seed(6186)
dt <- data.table(
id = rep(1:1000, times = 3),
query = rep(c("dolphin", "gorilla", "emu"), each = 1000),
pred_score = runif(3000, min = 0, max = 1)
)
dt[, relevant := FALSE]
dt[id %in% sample(1000, size = 25, replace = FALSE) & query == "dolphin", relevant := TRUE]
dt[id %in% sample(1000, size = 25, replace = FALSE) & query == "gorilla", relevant := TRUE]
dt[id %in% sample(1000, size = 25, replace = FALSE) & query == "emu", relevant := TRUE]
print(dt)
# id query pred_score relevant
# 1: 1 dolphin 0.3276352 FALSE
# 2: 2 dolphin 0.5503843 FALSE
# 3: 3 dolphin 0.2108938 FALSE
# 4: 4 dolphin 0.3693824 FALSE
# 5: 5 dolphin 0.5566705 FALSE
# ---
# 2996: 996 gorilla 0.7887701 FALSE
# 2997: 997 gorilla 0.1574011 FALSE
# 2998: 998 gorilla 0.2992015 FALSE
# 2999: 999 gorilla 0.3836474 FALSE
# 3000: 1000 gorilla 0.6554015 TRUE