Fix wrong assignment in computeMetricVariance for Frequency

[riemanli]

No description provided.

[wfa-reviewable]

This change is 

[riemanli]

Current dependencies on/for this PR:

• main
  • PR Fix wrong assignment in computeMetricVariance for Frequency #1464 👈

This stack of pull requests is managed by Graphite.

[jiayu-google]

Reviewable status: 0 of 2 files reviewed, 1 unresolved discussion (waiting on @chenweiw and @riemanli)

---

src/test/kotlin/org/wfanet/measurement/measurementconsumer/stats/VariancesTest.kt line 3354 at r1 (raw file):

    val maximumFrequency = 5
    val relativeFrequencyDistribution =
      (1..maximumFrequency).associateWith { (maximumFrequency - it) / 10.0 }

where is 10.0 from?

[chenweiw]

Line 3354: what does this line mean?

[chenweiw]

I'm trying to understand the number of the four lists in line 3388--3409. Where do these expected values come from?

[chenweiw]

So this addressed the issue that we discussed yesterday -- the stds of rk and rk+ estimations are always equal. Am I right?

[chenweiw]

Hi Rieman, I just made a couple of minor comments.

[riemanli]

Reviewable status: 0 of 2 files reviewed, 4 unresolved discussions (waiting on @chenweiw and @jiayu-google)

---

src/main/kotlin/org/wfanet/measurement/measurementconsumer/stats/Variances.kt line 630 at r1 (raw file):

Previously, chenweiw wrote…

So this addressed the issue that we discussed yesterday -- the stds of rk and rk+ estimations are always equal. Am I right?

Yes. The fix here is using the correct field in frequencyVariances

---

src/test/kotlin/org/wfanet/measurement/measurementconsumer/stats/VariancesTest.kt line 3354 at r1 (raw file):

Previously, jiayu-google wrote…

where is 10.0 from?

10.0 is the normalization term

---

src/test/kotlin/org/wfanet/measurement/measurementconsumer/stats/VariancesTest.kt line 3354 at r1 (raw file):

Previously, chenweiw wrote…

Line 3354: what does this line mean?

It creates a frequency distribution of [(5 - 1) / 10, (5 - 2) / 10, (5 - 3) / 10, (5 - 4) / 10, (5 - 5) / 10] = [0.4, 0.3, 0.2, 0.1, 0]

---

src/test/kotlin/org/wfanet/measurement/measurementconsumer/stats/VariancesTest.kt line 3412 at r1 (raw file):

Previously, chenweiw wrote…

I'm trying to understand the number of the four lists in line 3388--3409. Where do these expected values come from?

These values are generated from the Colab notebook you and Jiayu wrote. Basically, I use the notebook as a ground truth generator to ensure the code in prod works the same as the code in the Colab notebook.

[chenweiw]

Reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @jiayu-google)

[riemanli]

Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @jiayu-google and @Marco-Premier)

---

src/test/kotlin/org/wfanet/measurement/measurementconsumer/stats/VariancesTest.kt line 3354 at r1 (raw file):

Previously, riemanli (Rieman) wrote…

10.0 is the normalization term

It creates a frequency distribution of [(5 - 1) / 10, (5 - 2) / 10, (5 - 3) / 10, (5 - 4) / 10, (5 - 5) / 10] = [0.4, 0.3, 0.2, 0.1, 0]

[Marco-Premier]

Reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @jiayu-google and @stevenwarejones)

[stevenwarejones]

Reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @jiayu-google)

[riemanli]

Dismissed @jiayu-google from a discussion.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @jiayu-google)

[jiayu-google]

Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @riemanli)