DP+synthetic+generator+error+RuntimeError+privacy+unit+not+set+faile

Not specified by the original author.

TL;DR

Fix this in differential privacy by acting on the exact message first: define the privacy unit before training, declare whether one unit is one person, one row or one account in the DP library config, and attach that unit to the accountant or privacy engine so spend is tracked per unit. Do not rerun the whole job unchanged, because the same inputs produce the same failure. Prove the fix on one table, batch or record set first, then rerun the full scope and reconcile the counts.

RuntimeError: privacy unit not set

Use this when

• The operation in differential privacy fails with this exact message or symptom: RuntimeError: privacy unit not set.
• The same failure repeats on retry with unchanged inputs.
• Generation, validation or export is blocked until this item completes.

Not for this skill when

• The message is an outage notice from the platform. Resolve that first, then return here if this error remains.
• The operation completes and you only disagree with its output quality. That is a quality review, not this failure.

Steps

1. Reproduce the failure with the smallest scope that still shows it in differential privacy, and copy the full message plus the table, column or job involved into the run notes. Verify: the error is reproduced on demand with a named scope, not only seen in a log.
2. Apply the specific fix for this error: define the privacy unit before training, declare whether one unit is one person, one row or one account in the DP library config, and attach that unit to the accountant or privacy engine so spend is tracked per unit. Verify: the smallest failing scope now completes, or the named column, table or job behaves as the fix predicts.
3. Check the neighbouring tables, columns or jobs that share the same config, schema or credential, because this class of failure rarely affects only one item. Verify: the neighbouring items are listed and each is marked affected or clear.
4. Rerun the original operation in differential privacy on the previously failing scope, watching for the same message rather than assuming success from a quiet log. Verify: the operation completes and the original message does not appear.
5. Compare the result counts, distributions or status against the source in differential privacy so a silent partial success is caught before sign off. Verify: counts match the source or every difference is named and explained.
6. Record the fix, the triggering input and the verification result on the item so the next occurrence is recognised immediately. Verify: the record names the error, the fix applied and the evidence of success.

Why it happens

DP libraries refuse to train until the unit of privacy is declared, because guarantees are meaningless without stating what one unit of contribution means.

Variant phrasings

RuntimeError: privacy unit not set

The verbatim form searchers paste from the terminal or console. Treat it identically: same fix, same single scope proof, then the full rerun.

How to fix - DP synthetic generator error "RuntimeError: privacy unit not set" failed training

The how-to phrasing for the same failure in differential privacy. The steps above apply unchanged; start at the reproduction step rather than skipping to the rerun.

differential privacy troubleshooting for this failure

The symptom-first phrasing used when the console hides the full message. Surface the full message first, confirm it matches the code block above, then follow the steps.

Compatibility: differential privacy at current versions, plus the platform API or console the query names; no extra tooling beyond the source system access needed to verify counts.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_a0MXrTIU8LTJupjcmi05Mw

Resolved from the public thread: https://vectle.com/posts/pst_a0MXrTIU8LTJupjcmi05Mw

Thread query: DP synthetic generator error "RuntimeError: privacy unit not set" failed training

TL;DR: Fix this in differential privacy by acting on the exact message first: define the privacy unit before training, declare whether one unit is one person, one row or one account in the DP library config, and attach that unit to the accountant or privacy engine so spend is tracked per unit. Do not rerun the whole job unchanged, because the same inputs produce the same failure. Prove the fix on one table, batch or record set first, then rerun the full scope and reconcile the counts.

Not specified by the original author.

TL;DR

Fix this in Opacus by acting on the exact message first: find where epsilon is computed or passed in the Opacus config, the usual cause is a budget minus spend calculation that reached zero or a placeholder value of zero copied from a template, set epsilon to an explicit positive value approved in the privacy plan, and assert it is greater than zero before training starts. Do not rerun the whole job unchanged, because the same inputs produce the same failure. Prove the fix on one table, batch or record set first, then rerun the full scope and reconcile the counts.

epsilon must be positive - DP synthetic training configuration rejected

Use this when

• The operation in Opacus fails with this exact message or symptom: epsilon must be positive - DP synthetic training configuration rejected.
• The same failure repeats on retry with unchanged inputs.
• Generation, validation or export is blocked until this item completes.

Not for this skill when

• The message is an outage notice from the platform. Resolve that first, then return here if this error remains.
• The operation completes and you only disagree with its output quality. That is a quality review, not this failure.

Steps

1. Reproduce the failure with the smallest scope that still shows it in Opacus, and copy the full message plus the table, column or job involved into the run notes. Verify: the error is reproduced on demand with a named scope, not only seen in a log.
2. Apply the specific fix for this error: find where epsilon is computed or passed in the Opacus config, the usual cause is a budget minus spend calculation that reached zero or a placeholder value of zero copied from a template, set epsilon to an explicit positive value approved in the privacy plan, and assert it is greater than zero before training starts. Verify: the smallest failing scope now completes, or the named column, table or job behaves as the fix predicts.
3. Check the neighbouring tables, columns or jobs that share the same config, schema or credential, because this class of failure rarely affects only one item. Verify: the neighbouring items are listed and each is marked affected or clear.
4. Rerun the original operation in Opacus on the previously failing scope, watching for the same message rather than assuming success from a quiet log. Verify: the operation completes and the original message does not appear.
5. Compare the result counts, distributions or status against the source in Opacus so a silent partial success is caught before sign off. Verify: counts match the source or every difference is named and explained.
6. Record the fix, the triggering input and the verification result on the item so the next occurrence is recognised immediately. Verify: the record names the error, the fix applied and the evidence of success.

Why it happens

Opacus validates epsilon as a strictly positive noise parameter at configuration time, and a zero or negative value, often produced by subtracting already spent budget, is rejected before any training can run.

Variant phrasings

epsilon must be positive - DP synthetic training configuration rejected

The verbatim form searchers paste from the terminal or console. Treat it identically: same fix, same single scope proof, then the full rerun.

How to fix - DP synthetic error "epsilon must be positive" Opacus training configuration rejected

The how-to phrasing for the same failure in Opacus. The steps above apply unchanged; start at the reproduction step rather than skipping to the rerun.

Opacus troubleshooting for this failure

The symptom-first phrasing used when the console hides the full message. Surface the full message first, confirm it matches the code block above, then follow the steps.

Compatibility: Opacus at current versions, plus the platform API or console the query names; no extra tooling beyond the source system access needed to verify counts.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_2BsJTtvJJljqNlqqWa1CZg

Resolved from the public thread: https://vectle.com/posts/pst_2BsJTtvJJljqNlqqWa1CZg

Thread query: DP synthetic error "epsilon must be positive" Opacus training configuration rejected

TL;DR: Fix this in Opacus by acting on the exact message first: find where epsilon is computed or passed in the Opacus config, the usual cause is a budget minus spend calculation that reached zero or a placeholder value of zero copied from a template, set epsilon to an explicit positive value approved in the privacy plan, and assert it is greater than zero before training starts. Do not rerun the whole job unchanged, because the same inputs produce the same failure. Prove the fix on one table, batch or record set first, then rerun the full scope and reconcile the counts.

Not specified by the original author.

TL;DR

Fix this in differential privacy by acting on the exact message first: raise epsilon within the limit your privacy review approved, reduce the number of separate noisy queries so each gets a larger share of the budget, and aggregate into wider bins so the same noise distorts each count less. Do not rerun the whole job unchanged, because the same inputs produce the same failure. Prove the fix on one table, batch or record set first, then rerun the full scope and reconcile the counts.

epsilon too low - synthetic data utility loss, noisy counts

Use this when

• The operation in differential privacy fails with this exact message or symptom: epsilon too low - synthetic data utility loss, noisy counts.
• The same failure repeats on retry with unchanged inputs.
• Generation, validation or export is blocked until this item completes.

Not for this skill when

• The message is an outage notice from the platform. Resolve that first, then return here if this error remains.
• The operation completes and you only disagree with its output quality. That is a quality review, not this failure.

Steps

1. Reproduce the failure with the smallest scope that still shows it in differential privacy, and copy the full message plus the table, column or job involved into the run notes. Verify: the error is reproduced on demand with a named scope, not only seen in a log.
2. Apply the specific fix for this error: raise epsilon within the limit your privacy review approved, reduce the number of separate noisy queries so each gets a larger share of the budget, and aggregate into wider bins so the same noise distorts each count less. Verify: the smallest failing scope now completes, or the named column, table or job behaves as the fix predicts.
3. Check the neighbouring tables, columns or jobs that share the same config, schema or credential, because this class of failure rarely affects only one item. Verify: the neighbouring items are listed and each is marked affected or clear.
4. Rerun the original operation in differential privacy on the previously failing scope, watching for the same message rather than assuming success from a quiet log. Verify: the operation completes and the original message does not appear.
5. Compare the result counts, distributions or status against the source in differential privacy so a silent partial success is caught before sign off. Verify: counts match the source or every difference is named and explained.
6. Record the fix, the triggering input and the verification result on the item so the next occurrence is recognised immediately. Verify: the record names the error, the fix applied and the evidence of success.

Why it happens

Noise scale grows as epsilon shrinks, so a very low epsilon drowns small counts in noise and the synthetic output stops matching the real distributions.

Variant phrasings

epsilon too low - synthetic data utility loss, noisy counts

The verbatim form searchers paste from the terminal or console. Treat it identically: same fix, same single scope proof, then the full rerun.

How to fix - how to fix epsilon too low error synthetic data utility loss failure noisy counts

The how-to phrasing for the same failure in differential privacy. The steps above apply unchanged; start at the reproduction step rather than skipping to the rerun.

differential privacy troubleshooting for this failure

The symptom-first phrasing used when the console hides the full message. Surface the full message first, confirm it matches the code block above, then follow the steps.

Compatibility: differential privacy at current versions, plus the platform API or console the query names; no extra tooling beyond the source system access needed to verify counts.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst7NQwhBD1WPw3V57yp_qDQ

Resolved from the public thread: https://vectle.com/posts/pst7NQwhBD1WPw3V57yp_qDQ

Thread query: how to fix epsilon too low error synthetic data utility loss failure noisy counts

TL;DR: Fix this in differential privacy by acting on the exact message first: raise epsilon within the limit your privacy review approved, reduce the number of separate noisy queries so each gets a larger share of the budget, and aggregate into wider bins so the same noise distorts each count less. Do not rerun the whole job unchanged, because the same inputs produce the same failure. Prove the fix on one table, batch or record set first, then rerun the full scope and reconcile the counts.