67 files
This commit is contained in:
Home Assistant Version Control
@@ -137,6 +137,7 @@ CONF_ANTI_WRINKLE_MAX_DURATION = "anti_wrinkle_max_duration" # Seconds to treat
|
||||
CONF_ANTI_WRINKLE_EXIT_POWER = "anti_wrinkle_exit_power" # W threshold for true-off exit
|
||||
CONF_ANTI_WRINKLE_IDLE_TIMEOUT = "anti_wrinkle_idle_timeout" # Seconds below exit power before anti-wrinkle ends
|
||||
CONF_DISHWASHER_END_SPIKE_QUIET_RELEASE = "dishwasher_end_spike_quiet_release" # Dishwasher: sustained-quiet seconds after expected duration that release the end-of-cycle drain wait early (#379)
|
||||
CONF_SMART_TERMINATION_DURATION_RATIO = "smart_termination_duration_ratio" # Fraction of the matched profile's expected (mean) duration that Smart Termination requires before it may fire (#393)
|
||||
CONF_DELAY_START_DETECT_ENABLED = "delay_start_detect_enabled" # Enable delayed-start detection
|
||||
CONF_DELAY_CONFIRM_SECONDS = "delay_confirm_seconds" # Seconds power must stay in standby band before DELAY_WAIT engages
|
||||
CONF_DELAY_TIMEOUT_HOURS = "delay_timeout_hours" # Safety timeout (hours) while waiting to start
|
||||
@@ -297,7 +298,8 @@ DEFAULT_PROFILE_MATCH_MAX_DURATION_RATIO = 1.5
|
||||
DEFAULT_MAX_PAST_CYCLES = 200
|
||||
DEFAULT_MAX_FULL_TRACES_PER_PROFILE = 20
|
||||
DEFAULT_MAX_FULL_TRACES_UNLABELED = 20
|
||||
DEFAULT_WATCHDOG_INTERVAL = 30 # Derived: 2 * sampling_interval + 1
|
||||
DEFAULT_WATCHDOG_INTERVAL = 30 # Floor; effective default is resolved per device
|
||||
# as max(this, 2*sampling_interval + 1) - see resolve_watchdog_interval_default (#396).
|
||||
DEFAULT_MATCH_PERSISTENCE = 3
|
||||
DEFAULT_END_REPEAT_COUNT = 1 # 1 = current behavior (no repeat required)
|
||||
|
||||
@@ -487,6 +489,13 @@ DEFAULT_DTW_MODE = "ensemble"
|
||||
MATCH_DTW_RESAMPLE_N = 200 # common grid length for "scaled"/"ddtw" DTW
|
||||
MATCH_DDTW_DIST_SCALE = 30.0 # half-saturation for derivative-DTW distance
|
||||
MATCH_DTW_ENSEMBLE_W = 0.7 # weight on L1 vs DDTW in "ensemble" mode
|
||||
# Envelope alignment grid cap: maximum number of time-grid points used by
|
||||
# compute_envelope_worker. The DTW cost matrix is (n+1)x(m+1)x8 B float64;
|
||||
# both n and m derive from this cap, so memory is bounded to roughly
|
||||
# MAX_ALIGN_GRID_POINTS² x 8 B ≈ 32 MB at 2000 — regardless of cycle duration
|
||||
# or recording density. Without this cap a 4 h cycle at 1 Hz asks for 1.81 GB
|
||||
# in a single np.full, which OOM-kills Home Assistant (issue #388).
|
||||
MAX_ALIGN_GRID_POINTS = 2000
|
||||
# Ambiguity: top1-top2 score gap below this flags the match as ambiguous.
|
||||
MATCH_AMBIGUITY_MARGIN = 0.05
|
||||
# Smart Termination landscape guard: when a non-winning candidate is at least this
|
||||
@@ -499,6 +508,70 @@ MATCH_AMBIGUITY_MARGIN = 0.05
|
||||
SMART_TERM_LANDSCAPE_RATIO = 1.5 # candidate must be >= 1.5× the matched duration
|
||||
SMART_TERM_LANDSCAPE_MIN_SHAPE = 0.40 # minimum shape score (pre-Stage-4) to qualify
|
||||
|
||||
# Issue #364: the landscape guard above has three structural false negatives, all
|
||||
# reproduced by field reports on a multi-programme Miele washer:
|
||||
# (1) it needs a longer profile to EXIST in the candidate pool, so an untrained
|
||||
# longer programme is uncatchable;
|
||||
# (2) it qualifies the longer candidate on its shape score against its FULL
|
||||
# envelope, but a trace that is only part-way through a longer programme
|
||||
# scores poorly against that programme's whole curve;
|
||||
# (3) the 1.5 ratio is knife-edge - on a real 13-programme washer the observed
|
||||
# neighbour ratios are 1.12-1.48, so the guard never fires at all.
|
||||
#
|
||||
# Two independent additions, both shorten-only (they can only ever BLOCK an early
|
||||
# finish, never end a cycle sooner).
|
||||
#
|
||||
# (a) Prefix scoring (fixes 2 + 3). A longer candidate is re-scored against its own
|
||||
# curve TRUNCATED to the elapsed duration, which is an apples-to-apples comparison
|
||||
# and lands on the same 0-1 scale as `shape_score` (same find_best_alignment, same
|
||||
# DTW blend). Because it compares equal-length series over the whole overlap it
|
||||
# reads systematically higher than the full-envelope score, so it gets its OWN
|
||||
# threshold rather than reusing SMART_TERM_LANDSCAPE_MIN_SHAPE. The load-bearing
|
||||
# term is the MARGIN over the winner ("the longer programme explains this trace at
|
||||
# least this much better than the short one does"), which is scale-free; the floor
|
||||
# only rejects candidates that fit nothing. Measured on 20 real cycles + 7
|
||||
# envelopes (37 prefix-cut positives vs 17 genuine-cycle negatives): margin 0.15
|
||||
# catches 26/37 splits for 1/17 false blocks, while simply lowering the ratio to
|
||||
# 1.35/1.15 costs 2/17 and 4/17 false blocks for no measured gain.
|
||||
SMART_TERM_PREFIX_MARGIN = 0.15 # prefix score must beat the winner by this
|
||||
SMART_TERM_PREFIX_MIN_SHAPE = 0.40 # absolute floor on the prefix score
|
||||
SMART_TERM_PREFIX_MIN_RATIO = 1.10 # noise guard: ignore near-equal durations
|
||||
SMART_TERM_PREFIX_MAX_CANDIDATES = 3 # cap prefix scorings per match (cost control)
|
||||
SMART_TERM_PREFIX_MIN_POINTS = 12 # mirrors the matcher's >=12-sample floor
|
||||
SMART_TERM_PREFIX_MIN_COVERAGE = 0.90 # template span must cover >=90% of its duration
|
||||
|
||||
# (b) Power plausibility (fixes 1, the untrained case, which no candidate-pool guard
|
||||
# can reach). Both Smart-Termination paths key on `elapsed >= 0.98 * expected` and
|
||||
# neither checks whether the appliance is still WORKING, so a mis-matched shorter
|
||||
# profile finalises a cycle mid-wash. Compare the trailing mean power against what
|
||||
# the matched profile itself draws at its own end: if we are drawing several times
|
||||
# that, this is not the end of anything.
|
||||
#
|
||||
# The two windows must cover the same FRACTION of the run, or the comparison is not
|
||||
# like-for-like. A fixed 300 s trailing window is 4% of a cotton wash but a third
|
||||
# of a 15-minute "Spin & Drain", whose trailing mean is then the spin itself while
|
||||
# its profile tail is the quiet moment after the pump stops - ratios of 45-310x on
|
||||
# perfectly normal cycle ends. So the trailing window is
|
||||
# `expected_duration * SMART_TERM_TAIL_WINDOW_FRAC`, clamped; that alone is strictly
|
||||
# better at every threshold (e.g. at 4.0x: false blocks 5% -> 3%).
|
||||
#
|
||||
# Swept with devtools/prefix_guard_eval.py over the whole cycle_data corpus (19
|
||||
# devices, 225 labelled cycles, leave-one-out): 114 folds where a shorter profile
|
||||
# is winning mid-cycle (the #364 split condition) vs 165 genuine cycle ends.
|
||||
# ratio caught false-blocked
|
||||
# 3.0 27% 8%
|
||||
# 3.5 25% 4% <- shipped, the knee
|
||||
# 4.0 20% 3%
|
||||
# 3.0 -> 3.5 halves the false blocks for 2pp of catch, and the reported cases sit
|
||||
# at 4-8x so they stay caught. A false block only costs a later finish (the
|
||||
# power-based fallback timeout still ends the cycle); a miss costs a split cycle.
|
||||
# ~1 in 5 of the remaining false blocks had a wrong top-1 anyway, where blocking is right.
|
||||
SMART_TERM_TAIL_MAX_RATIO = 3.5 # block while trailing mean > this x profile tail
|
||||
SMART_TERM_TAIL_WINDOW_S = 300.0 # upper clamp on the trailing window
|
||||
SMART_TERM_TAIL_WINDOW_MIN_S = 60.0 # lower clamp (short programmes)
|
||||
SMART_TERM_TAIL_MIN_POINTS = 3 # too few samples -> no opinion, do not block
|
||||
SMART_TERM_TAIL_WINDOW_FRAC = 0.05 # both windows = last 5% of the run
|
||||
|
||||
# Number of points in the compact reference-profile curve exposed on the
|
||||
# `_program` sensor (`profile_store.reference_curve`). Chosen so the resulting
|
||||
# `[[offset_s, watts], ...]` attribute stays comfortably under ~1 KB regardless
|
||||
@@ -773,6 +846,20 @@ STARTING_PAUSED_TRUE_OFF_TIMEOUT_SECONDS = 300.0
|
||||
# a washer's longest legitimate mid-cycle soak trough while capping the pathology.
|
||||
WASHER_SMART_TERMINATION_DEBOUNCE_MAX_SECONDS = 600.0
|
||||
|
||||
# Fraction of the matched profile's expected (mean) duration that Smart Termination
|
||||
# requires before it may fire (#393). self._expected_duration is the profile's
|
||||
# outlier-filtered ARITHMETIC MEAN, so a fixed 0.98 gate against a mean is
|
||||
# structurally unreachable for appliances whose runtime depends on load, fill level
|
||||
# or inlet temperature - about half of those cycles are shorter than their own mean
|
||||
# by construction and can never take the fast path. Exposed as the per-device
|
||||
# CONF_SMART_TERMINATION_DURATION_RATIO option (range 0.50-1.00; empty = default).
|
||||
# The default is device-type-resolved: dishwashers keep the conservative 0.99
|
||||
# (their programs are fixed, so the spread is small) while everything else keeps
|
||||
# 0.98. The dishwasher pump-out relief (0.90 once the terminal pump-out spike is
|
||||
# confirmed) is combined with the configured value via min(), so the option can
|
||||
# only ever LOOSEN the gate, never tighten it.
|
||||
DEFAULT_SMART_TERMINATION_DURATION_RATIO = 0.98
|
||||
|
||||
DEFAULT_OFF_DELAY_BY_DEVICE = {
|
||||
DEVICE_TYPE_DISHWASHER: 1800, # 30 min (Drying)
|
||||
DEVICE_TYPE_BREAD_MAKER: 300, # 5 min (Keep-warm phase after baking)
|
||||
@@ -842,11 +929,71 @@ DEFAULT_SAMPLING_INTERVAL_BY_DEVICE = {
|
||||
DEVICE_TYPE_PUMP: 10.0, # 10s - pump cycles can be <30 s; 30s default would miss them
|
||||
}
|
||||
|
||||
|
||||
def resolve_sampling_interval_default(device_type: str) -> float:
|
||||
"""Device-resolved default sampling interval (#396).
|
||||
|
||||
Single source of truth for the sampling default, so the manager, the panel
|
||||
(via ws_get_options) and the config migration all agree. Wet appliances
|
||||
sample fast (2 s) to capture the rapid 0<->150 W oscillation; everything else
|
||||
keeps the coarse 30 s scalar.
|
||||
"""
|
||||
return DEFAULT_SAMPLING_INTERVAL_BY_DEVICE.get(device_type, DEFAULT_SAMPLING_INTERVAL)
|
||||
|
||||
|
||||
def resolve_watchdog_interval_default(device_type: str) -> int:
|
||||
"""Device-resolved watchdog tick default (#396).
|
||||
|
||||
The panel enforces watchdog_interval >= 2*sampling_interval (a publish-on-change
|
||||
sensor can skip a sample, so the staleness tick must be coarser than the
|
||||
sampling gap). Derived as max(DEFAULT_WATCHDOG_INTERVAL, 2*sampling+1): 30 for
|
||||
the fast/pump types (30 already clears 2*2 / 2*10), 61 for the 30 s-sampling
|
||||
types. Never smaller than the 30 s floor so a fast-sampling device does not get
|
||||
an over-aggressive watchdog.
|
||||
"""
|
||||
sampling = resolve_sampling_interval_default(device_type)
|
||||
return int(max(DEFAULT_WATCHDOG_INTERVAL, 2.0 * sampling + 1.0))
|
||||
|
||||
|
||||
def resolve_start_duration_default(device_type: str) -> float:
|
||||
"""Device-resolved start-debounce default (#396).
|
||||
|
||||
The panel enforces start_duration_threshold >= sampling_interval (a debounce
|
||||
shorter than one sample lets a single spike open a cycle). Derived as
|
||||
max(DEFAULT_START_DURATION_THRESHOLD, sampling): 5 s for the fast types, the
|
||||
sampling interval for the coarser ones.
|
||||
"""
|
||||
sampling = resolve_sampling_interval_default(device_type)
|
||||
return max(DEFAULT_START_DURATION_THRESHOLD, sampling)
|
||||
|
||||
|
||||
# Default profile match min duration ratio by device type
|
||||
DEFAULT_PROFILE_MATCH_MIN_DURATION_RATIO_BY_DEVICE = {
|
||||
DEVICE_TYPE_DISHWASHER: 0.10,
|
||||
}
|
||||
|
||||
# Default Smart-Termination duration ratio by device type (#393). Dishwashers run
|
||||
# fixed programs (measured spread +4%/+17% around the mean), so the conservative
|
||||
# 0.99 gate is defensible there; every other type keeps the scalar
|
||||
# DEFAULT_SMART_TERMINATION_DURATION_RATIO (0.98). Resolved in the config builder,
|
||||
# never in the gate, so playground.effective_settings() always sees a real float.
|
||||
DEFAULT_SMART_TERMINATION_DURATION_RATIO_BY_DEVICE = {
|
||||
DEVICE_TYPE_DISHWASHER: 0.99,
|
||||
}
|
||||
|
||||
|
||||
def resolve_smart_termination_duration_ratio_default(device_type: str) -> float:
|
||||
"""Device-resolved Smart-Termination duration ratio default (#393).
|
||||
|
||||
Single source of truth shared by the manager (config build/reload), the
|
||||
Playground fallback config and the panel (via ws_get_options), so the value
|
||||
the panel pre-populates always matches the one the detector actually uses:
|
||||
0.99 for dishwashers (fixed programs), 0.98 for everything else.
|
||||
"""
|
||||
return DEFAULT_SMART_TERMINATION_DURATION_RATIO_BY_DEVICE.get(
|
||||
device_type, DEFAULT_SMART_TERMINATION_DURATION_RATIO
|
||||
)
|
||||
|
||||
# Profile groups (Stage 5): the matcher only collapses a group into one
|
||||
# aggregate candidate when its members' minimum pairwise shape similarity is at
|
||||
# least this. Similarity is DTW/Sakoe-Chiba on peak-normalised envelopes, so it
|
||||
@@ -872,9 +1019,24 @@ GROUP_MIN_COHESION = 0.80
|
||||
# v11 is a marker-only bump: per-phase profiles (envelope["phase_profile"]) are
|
||||
# derived cache populated by async_rebuild_envelope, so no data migration is
|
||||
# needed - they self-populate on the next envelope rebuild.
|
||||
STORAGE_VERSION = 11
|
||||
# v12: initialize `backfill_cycles`, the third cycle list (issue #344). Cycles
|
||||
# recovered from raw power history predating the integration are auto-detected and
|
||||
# unverified, so they belong in neither `past_cycles` (which feeds lifetime stats, ML
|
||||
# training labels and the feedback queue, and is retention-evicted oldest-first) nor
|
||||
# `reference_cycles` (curated community-store templates, golden by construction).
|
||||
# Additive `setdefault`, so it is idempotent and loses nothing.
|
||||
STORAGE_VERSION = 12
|
||||
STORAGE_KEY = "ha_washdata"
|
||||
|
||||
# ─── Config-entry schema version (NOT the storage version above) ───────────────
|
||||
# Single source for the config-entry schema: `ConfigFlow.VERSION`/`MINOR_VERSION`, every
|
||||
# stepwise block in `async_migrate_entry`, and the `minor_version=` the one-pass legacy
|
||||
# migration writes all read from here. They must move together - a bump that misses one
|
||||
# leaves an entry a version short, which then re-migrates on every start - and repeating
|
||||
# the literals in three places is what made that easy to do.
|
||||
CONFIG_ENTRY_VERSION = 3
|
||||
CONFIG_ENTRY_MINOR_VERSION = 10
|
||||
|
||||
# Notification events
|
||||
EVENT_CYCLE_STARTED = "ha_washdata_cycle_started"
|
||||
EVENT_CYCLE_ENDED = "ha_washdata_cycle_ended"
|
||||
@@ -1066,3 +1228,80 @@ PLAYGROUND_STRESS_MAX_IDLE_W: float = 100000.0 # upper bound for a manual
|
||||
# value to entry.options is always an explicit, per-setting user action.
|
||||
PLAYGROUND_PRESET_MAX: int = 30 # per-device cap (keeps the store small)
|
||||
PLAYGROUND_PRESET_NAME_MAX: int = 60 # preset name length cap
|
||||
|
||||
# ─── Which cycle categories count as evidence for a profile ────────────────────
|
||||
# A profile's envelope (its average curve + duration/energy spread) and the matching
|
||||
# template are built from stored cycles. By default all three categories count. Untick a
|
||||
# category and it stops shaping profiles - useful when you do not trust imported data -
|
||||
# without deleting anything: the cycles remain stored, listed and deletable.
|
||||
#
|
||||
# This gates *evidence* only (`ProfileStore.iter_evidence_cycles`), never
|
||||
# `iter_stored_cycles`/`find_stored_cycle`. Profile garbage collection and sample repair
|
||||
# delete or re-point a profile whose sample cycle resolves to nothing, so they must keep
|
||||
# seeing every stored cycle: a cycle excluded from evidence is still a stored cycle, and
|
||||
# gating those lookups would destroy a backfill-only profile the moment someone unticked
|
||||
# imported history.
|
||||
CONF_PROFILE_EVIDENCE_SOURCES = "profile_evidence_sources"
|
||||
EVIDENCE_REAL_CYCLES = "real_cycles"
|
||||
EVIDENCE_REFERENCE_CYCLES = "reference_cycles"
|
||||
EVIDENCE_BACKFILL_CYCLES = "backfill_cycles"
|
||||
# `real_cycles`/`reference_cycles` match the export taxonomy (`_EXPORT_CATEGORIES`); the
|
||||
# evidence view adds `backfill_cycles`, which the selective-export wizard does not yet
|
||||
# enumerate (whole-store export still round-trips it).
|
||||
PROFILE_EVIDENCE_SOURCES = (
|
||||
EVIDENCE_REAL_CYCLES,
|
||||
EVIDENCE_REFERENCE_CYCLES,
|
||||
EVIDENCE_BACKFILL_CYCLES,
|
||||
)
|
||||
# All three: the pre-setting behaviour, so an upgrade changes nothing.
|
||||
DEFAULT_PROFILE_EVIDENCE_SOURCES = list(PROFILE_EVIDENCE_SOURCES)
|
||||
|
||||
# ─── Historical power-data import (issue #344) ─────────────────────────────────
|
||||
# An HA history export (or a recorder read) is a *change-based* stream: a steady
|
||||
# 0 W emits no rows at all, so it cannot be fed to the detector as-is (doing so
|
||||
# produces multi-day `force_stopped` blobs). `history_import.py` pre-segments the
|
||||
# stream into activity blocks first; these constants govern that pre-pass.
|
||||
HISTORY_IMPORT_MAX_BYTES: int = 32 * 1024 * 1024 # staged upload cap (~32 MiB of CSV text)
|
||||
HISTORY_IMPORT_MAX_ROWS: int = 500_000 # parsed-row cap (≈ a month at 5 s)
|
||||
HISTORY_IMPORT_CHUNK_BYTES: int = 512 * 1024 # per-WS-message upload chunk (frame cap is 4 MiB)
|
||||
HISTORY_IMPORT_CHUNK_SAMPLES: int = 4000 # samples replayed per executor job
|
||||
HISTORY_IMPORT_MIN_BLOCK_SAMPLES: int = 20 # floor for the per-block sample gate
|
||||
HISTORY_IMPORT_MAX_MEDIAN_INTERVAL_S: float = 120.0 # floor for the per-block cadence gate; the
|
||||
# effective gate is
|
||||
# max(this, 4 x sampling_interval) so a plug
|
||||
# that legitimately reports every 60 s is not
|
||||
# rejected
|
||||
HISTORY_IMPORT_EDGE_GAP_S: float = 60.0 # leading samples this far from the block body
|
||||
# are hourly-average debris and are trimmed
|
||||
# (leading edge ONLY - trimming the trailing
|
||||
# edge eats a real cycle's low-power tail)
|
||||
HISTORY_IMPORT_MAX_BLOCK_SPAN_S: float = 12 * 3600.0 # a block longer than this can only produce the
|
||||
# detector's 8 h `force_stopped` blob, so it is
|
||||
# reported rather than replayed
|
||||
HISTORY_IMPORT_DENSIFY_STEP_S: float = 30.0 # cadence of the synthetic samples inserted into
|
||||
# a carried-forward *quiet* gap, so the
|
||||
# detector's gap-free quiet tally can accrue
|
||||
# exactly as it does live
|
||||
HISTORY_IMPORT_TAIL_STEP_S: float = 30.0 # synthetic quiet-tail cadence used to close the
|
||||
# last cycle of a block
|
||||
HISTORY_IMPORT_MAX_SEGMENTS: int = 60 # candidates surfaced by one scan
|
||||
HISTORY_IMPORT_MAX_TOTAL_CYCLES: int = 200 # total backfilled cycles kept per device
|
||||
# (`backfill_cycles` has no retention pass and
|
||||
# the whole store blob is rewritten on every
|
||||
# throttled active-cycle save)
|
||||
HISTORY_IMPORT_RECORDER_MAX_DAYS: int = 3700 # ~10 years. HA's default `purge_keep_days`
|
||||
# is 10, but a recorder configured to keep
|
||||
# full-resolution states for years is a real
|
||||
# setup and must not be capped out of reach.
|
||||
# Reaching past what the recorder holds simply
|
||||
# returns fewer rows; the real guard is
|
||||
# HISTORY_IMPORT_MAX_ROWS, which stops the
|
||||
# day-by-day read as soon as enough accrues.
|
||||
HISTORY_IMPORT_RECORDER_EMPTY_DAY_STOP: int = 30 # consecutive empty days that end the walk.
|
||||
# Within the retention window a day always
|
||||
# yields at least the carried start-time state,
|
||||
# so a run of truly empty days means the
|
||||
# recorder has been purged past this point -
|
||||
# without this, a 10-year request would issue
|
||||
# thousands of pointless queries.
|
||||
HISTORY_IMPORT_SOURCE: str = "history_import" # `meta.source` marker on imported cycles
|
||||
|
||||
Reference in New Issue
Block a user