tweak

esphome/components/sensor/__init__.py

@@ -249,6 +249,9 @@ MaxFilter = sensor_ns.class_("MaxFilter", Filter)
 SlidingWindowMovingAverageFilter = sensor_ns.class_(
     "SlidingWindowMovingAverageFilter", Filter
 )
+StreamingMinFilter = sensor_ns.class_("StreamingMinFilter", Filter)
+StreamingMaxFilter = sensor_ns.class_("StreamingMaxFilter", Filter)
+StreamingMovingAverageFilter = sensor_ns.class_("StreamingMovingAverageFilter", Filter)
 ExponentialMovingAverageFilter = sensor_ns.class_(
     "ExponentialMovingAverageFilter", Filter
 )
@@ -452,6 +455,19 @@ async def skip_initial_filter_to_code(config, filter_id):
 
 @FILTER_REGISTRY.register("min", MinFilter, MIN_SCHEMA)
 async def min_filter_to_code(config, filter_id):
+    window_size = config[CONF_WINDOW_SIZE]
+    send_every = config[CONF_SEND_EVERY]
+    send_first_at = config[CONF_SEND_FIRST_AT]
+
+    # Optimization: Use streaming filter for batch windows (window_size == send_every)
+    # Saves 99.98% memory for large windows (e.g., 20KB → 4 bytes for window_size=5000)
+    if window_size == send_every:
+        return cg.new_Pvariable(
+            filter_id,
+            StreamingMinFilter,
+            window_size,
+            send_first_at,
+        )
     return cg.new_Pvariable(
         filter_id,
         config[CONF_WINDOW_SIZE],
@@ -474,6 +490,19 @@ MAX_SCHEMA = cv.All(
 
 @FILTER_REGISTRY.register("max", MaxFilter, MAX_SCHEMA)
 async def max_filter_to_code(config, filter_id):
+    window_size = config[CONF_WINDOW_SIZE]
+    send_every = config[CONF_SEND_EVERY]
+    send_first_at = config[CONF_SEND_FIRST_AT]
+
+    # Optimization: Use streaming filter for batch windows (window_size == send_every)
+    # Saves 99.98% memory for large windows (e.g., 20KB → 4 bytes for window_size=5000)
+    if window_size == send_every:
+        return cg.new_Pvariable(
+            filter_id,
+            StreamingMaxFilter,
+            window_size,
+            send_first_at,
+        )
     return cg.new_Pvariable(
         filter_id,
         config[CONF_WINDOW_SIZE],
@@ -500,6 +529,19 @@ SLIDING_AVERAGE_SCHEMA = cv.All(
     SLIDING_AVERAGE_SCHEMA,
 )
 async def sliding_window_moving_average_filter_to_code(config, filter_id):
+    window_size = config[CONF_WINDOW_SIZE]
+    send_every = config[CONF_SEND_EVERY]
+    send_first_at = config[CONF_SEND_FIRST_AT]
+
+    # Optimization: Use streaming filter for batch windows (window_size == send_every)
+    # Saves 99.94% memory for large windows (e.g., 20KB → 12 bytes for window_size=5000)
+    if window_size == send_every:
+        return cg.new_Pvariable(
+            filter_id,
+            StreamingMovingAverageFilter,
+            window_size,
+            send_first_at,
+        )
     return cg.new_Pvariable(
         filter_id,
         config[CONF_WINDOW_SIZE],

esphome/components/sensor/filter.cpp

@@ -468,5 +468,78 @@ optional<float> ToNTCTemperatureFilter::new_value(float value) {
   return temp;
 }
 
+// StreamingFilter (base class)
+StreamingFilter::StreamingFilter(size_t window_size, size_t send_first_at)
+    : window_size_(window_size), send_first_at_(send_first_at) {}
+
+optional<float> StreamingFilter::new_value(float value) {
+  // Process the value (child class tracks min/max/sum/etc)
+  this->process_value(value);
+
+  this->count_++;
+
+  // Check if we should send (handle send_first_at for first value)
+  bool should_send = false;
+  if (this->first_send_ && this->count_ >= this->send_first_at_) {
+    should_send = true;
+    this->first_send_ = false;
+  } else if (!this->first_send_ && this->count_ >= this->window_size_) {
+    should_send = true;
+  }
+
+  if (should_send) {
+    float result = this->compute_batch_result();
+    // Reset for next batch
+    this->count_ = 0;
+    this->reset_batch();
+    ESP_LOGVV(TAG, "StreamingFilter(%p)::new_value(%f) SENDING %f", this, value, result);
+    return result;
+  }
+
+  return {};
+}
+
+// StreamingMinFilter
+void StreamingMinFilter::process_value(float value) {
+  // Update running minimum (ignore NaN values)
+  if (!std::isnan(value)) {
+    this->current_min_ = std::isnan(this->current_min_) ? value : std::min(this->current_min_, value);
+  }
+}
+
+float StreamingMinFilter::compute_batch_result() { return this->current_min_; }
+
+void StreamingMinFilter::reset_batch() { this->current_min_ = NAN; }
+
+// StreamingMaxFilter
+void StreamingMaxFilter::process_value(float value) {
+  // Update running maximum (ignore NaN values)
+  if (!std::isnan(value)) {
+    this->current_max_ = std::isnan(this->current_max_) ? value : std::max(this->current_max_, value);
+  }
+}
+
+float StreamingMaxFilter::compute_batch_result() { return this->current_max_; }
+
+void StreamingMaxFilter::reset_batch() { this->current_max_ = NAN; }
+
+// StreamingMovingAverageFilter
+void StreamingMovingAverageFilter::process_value(float value) {
+  // Accumulate sum (ignore NaN values)
+  if (!std::isnan(value)) {
+    this->sum_ += value;
+    this->valid_count_++;
+  }
+}
+
+float StreamingMovingAverageFilter::compute_batch_result() {
+  return this->valid_count_ > 0 ? this->sum_ / this->valid_count_ : NAN;
+}
+
+void StreamingMovingAverageFilter::reset_batch() {
+  this->sum_ = 0.0f;
+  this->valid_count_ = 0;
+}
+
 }  // namespace sensor
 }  // namespace esphome

esphome/components/sensor/filter.h

@@ -504,5 +504,81 @@ class ToNTCTemperatureFilter : public Filter {
   double c_;
 };
 
+/** Base class for streaming filters (batch windows where window_size == send_every).
+ *
+ * When window_size equals send_every, we don't need a sliding window.
+ * This base class handles the common batching logic.
+ */
+class StreamingFilter : public Filter {
+ public:
+  StreamingFilter(size_t window_size, size_t send_first_at);
+
+  optional<float> new_value(float value) final;
+
+ protected:
+  /// Called by new_value() to process each value in the batch
+  virtual void process_value(float value) = 0;
+
+  /// Called by new_value() to compute the result after collecting window_size values
+  virtual float compute_batch_result() = 0;
+
+  /// Called by new_value() to reset internal state after sending a result
+  virtual void reset_batch() = 0;
+
+  size_t window_size_;
+  size_t count_{0};
+  size_t send_first_at_;
+  bool first_send_{true};
+};
+
+/** Streaming min filter for batch windows (window_size == send_every).
+ *
+ * Uses O(1) memory instead of O(n) by tracking only the minimum value.
+ */
+class StreamingMinFilter : public StreamingFilter {
+ public:
+  using StreamingFilter::StreamingFilter;
+
+ protected:
+  void process_value(float value) override;
+  float compute_batch_result() override;
+  void reset_batch() override;
+
+  float current_min_{NAN};
+};
+
+/** Streaming max filter for batch windows (window_size == send_every).
+ *
+ * Uses O(1) memory instead of O(n) by tracking only the maximum value.
+ */
+class StreamingMaxFilter : public StreamingFilter {
+ public:
+  using StreamingFilter::StreamingFilter;
+
+ protected:
+  void process_value(float value) override;
+  float compute_batch_result() override;
+  void reset_batch() override;
+
+  float current_max_{NAN};
+};
+
+/** Streaming moving average filter for batch windows (window_size == send_every).
+ *
+ * Uses O(1) memory instead of O(n) by tracking only sum and count.
+ */
+class StreamingMovingAverageFilter : public StreamingFilter {
+ public:
+  using StreamingFilter::StreamingFilter;
+
+ protected:
+  void process_value(float value) override;
+  float compute_batch_result() override;
+  void reset_batch() override;
+
+  float sum_{0.0f};
+  size_t valid_count_{0};
+};
+
 }  // namespace sensor
 }  // namespace esphome