update generation abstract model

Author: LongxingTan

CHANGELOG.md

@@ -1,5 +1,6 @@
 # Release notes
 
+
 ## v0.0.22
 
 - Add typed spatial arrangements and topology inputs to ``TimeSeriesBatch``.
@@ -8,6 +9,18 @@
 - Preserve shared graph metadata across Keras/``tf.data`` boundaries without
   copying static topology into every training window.
 - Add documented graph builders and a runnable spatial forecasting example.
+- Introduce a shared time-axis generation pipeline, typed `GenStep` hooks,
+  parameter and value processors, stopping criteria, and sample aggregation.
+- Resolve decoder capabilities when task models are built. RNN forecasts now
+  honor declared padding masks for variable-length histories.
+- **Checkpoint break:** Transformer causal cached decoding uses decoder format 2.
+  Legacy decoder checkpoints require migration or retraining; config-based
+  loaders reject missing or incompatible format markers before loading weights.
+  This change must ship as a minor release (at least 0.1.0).
+- Custom rollout strategies are runtime instances; serialized configs support
+  only the implemented built-in strategies. Remove unused generation quantiles.
+- Use `PointSampler` (`sampler="point"`), `TeacherForcingPolicy`, and
+  `TimeAxisEngine`.
 
 
 ## v.0.0.15

Makefile

@@ -1,4 +1,4 @@
-.PHONY: style test test-warnings docs pre-release help
+.PHONY: style test test-coverage test-warnings docs pre-release help
 
 # Directories to run style checks on
 CHECK_DIRS := tfts examples tests
@@ -21,6 +21,10 @@ TEST_ENV := TF_ENABLE_ONEDNN_OPTS=0 TF_CPP_MIN_LOG_LEVEL=3 PYTHONWARNINGS=ignore
 test:  ## Run unit tests without routine TensorFlow/Python warnings
 	$(TEST_ENV) $(UV_RUN) python -m unittest discover
 
+test-coverage:  ## Run unit tests and report branch coverage for the package
+	$(TEST_ENV) $(UV_RUN) coverage run --branch --source=tfts -m unittest discover
+	$(UV_RUN) coverage report -m
+
 test-warnings:  ## Run unit tests with all warnings and TensorFlow info logs
 	$(UV_RUN) python -m unittest discover
 

README.md

@@ -65,7 +65,7 @@ predict_sequence_length = 8
 
 model_name_or_path = 'seq2seq'  # 'wavenet', 'transformer', 'rnn', 'tcn', 'bert', 'dlinear', 'nbeats', 'informer', 'autoformer'
 config = AutoConfig.for_model(model_name_or_path)
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train(
     (x_train, y_train),
@@ -80,7 +80,7 @@ trainer.save_model(model_dir)
 
 # Load the model
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=x_valid[:1])
-pred = restored_model(x_valid, training=False).numpy()
+pred = restored_model(x_valid, training=False).predictions.numpy()
 trainer.plot(history=x_valid, true=y_valid, pred=pred)
 plt.show()
 ```
@@ -109,14 +109,14 @@ x_valid = np.random.rand(1, train_length, n_feature)
 y_valid = np.random.rand(1, predict_sequence_length, 1)
 
 config = AutoConfig.for_model('rnn')
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train(train_dataset=(x_train, y_train), valid_dataset=(x_valid, y_valid), epochs=1)
 
 model_dir = "./outputs/encoder_only"
 trainer.save_model(model_dir)
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=x_valid[:1])
-inference = restored_model(x_valid[:1], training=False).numpy()
+inference = restored_model(x_valid[:1], training=False).predictions.numpy()
 print(inference.shape)
 ```
 
@@ -147,14 +147,14 @@ x_valid = {
 y_valid = np.random.rand(1, predict_sequence_length, 1)
 
 config = AutoConfig.for_model("seq2seq")
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train((x_train, y_train), (x_valid, y_valid), epochs=1)
 
 model_dir = "./outputs/encoder_decoder"
 trainer.save_model(model_dir)
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=x_valid)
-inference = restored_model(x_valid, training=False).numpy()
+inference = restored_model(x_valid, training=False).predictions.numpy()
 print(inference.shape)
 ```
 
@@ -203,15 +203,15 @@ valid_loader = tf.data.Dataset.from_generator(
 valid_loader = valid_loader.batch(batch_size=1)
 
 config = AutoConfig.for_model("seq2seq")
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train(train_dataset=train_loader, valid_dataset=valid_loader, epochs=1)
 
 model_dir = "./outputs/encoder_decoder_tfdata"
 trainer.save_model(model_dir)
 inference_inputs = next(iter(valid_loader.take(1)))[0]
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=inference_inputs)
-inference = restored_model(inference_inputs, training=False).numpy()
+inference = restored_model(inference_inputs, training=False).predictions.numpy()
 print(inference.shape)
 ```
 
@@ -224,7 +224,7 @@ config = AutoConfig.for_model('rnn')
 print(config)
 config.rnn_hidden_size = 128
 
-model = AutoModelForForecasting.from_config(config, prediction_length=7)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=7)
 ```
 
 **Build your own model**
@@ -259,12 +259,18 @@ num_train_features = 15
 predict_sequence_length = 8
 
 def build_model():
-    inputs = Input([train_length, num_train_features])
-    config = AutoConfig.for_model("seq2seq")
-    backbone = AutoBackbone.from_config(config, prediction_length=predict_sequence_length)
-    outputs = backbone(inputs)
-    outputs = Dense(1, activation="sigmoid")(outputs)
-    model = tf.keras.Model(inputs=inputs, outputs=outputs)
+    class CustomForecaster(tf.keras.Model):
+        def __init__(self):
+            super().__init__()
+            config = AutoConfig.for_model("seq2seq")
+            self.backbone = AutoBackbone.from_config(config, output_chunk_length=predict_sequence_length)
+            self.projection = Dense(1, activation="sigmoid")
+
+        def call(self, inputs, training=None):
+            output = self.backbone(inputs, training=training)
+            return self.projection(output.native_forecast)
+
+    model = CustomForecaster()
     model.compile(loss="mse", optimizer="rmsprop")
     return model
 ```

README_CN.md

@@ -68,7 +68,7 @@ predict_sequence_length = 8
 
 model_name_or_path = 'seq2seq'  # 'wavenet', 'transformer', 'rnn', 'tcn', 'bert', 'dlinear', 'nbeats', 'informer', 'autoformer'
 config = AutoConfig.for_model(model_name_or_path)
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train(
     (x_train, y_train),
@@ -80,7 +80,7 @@ trainer.train(
 model_dir = "./outputs/quickstart_forecasting"
 trainer.save_model(model_dir)
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=x_valid[:1])
-pred = restored_model(x_valid, training=False).numpy()
+pred = restored_model(x_valid, training=False).predictions.numpy()
 trainer.plot(history=x_valid, true=y_valid, pred=pred)
 plt.show()
 ```
@@ -108,14 +108,14 @@ x_valid = np.random.rand(1, train_length, n_feature)
 y_valid = np.random.rand(1, predict_sequence_length, 1)
 
 config = AutoConfig.for_model("rnn")
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train(train_dataset=(x_train, y_train), valid_dataset=(x_valid, y_valid), epochs=1)
 
 model_dir = "./outputs/encoder_only"
 trainer.save_model(model_dir)
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=x_valid[:1])
-inference = restored_model(x_valid[:1], training=False).numpy()
+inference = restored_model(x_valid[:1], training=False).predictions.numpy()
 print(inference.shape)
 ```
 
@@ -145,14 +145,14 @@ x_valid = {
 y_valid = np.random.rand(1, predict_sequence_length, 1)
 
 config = AutoConfig.for_model("seq2seq")
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train((x_train, y_train), (x_valid, y_valid), epochs=1)
 
 model_dir = "./outputs/encoder_decoder"
 trainer.save_model(model_dir)
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=x_valid)
-inference = restored_model(x_valid, training=False).numpy()
+inference = restored_model(x_valid, training=False).predictions.numpy()
 print(inference.shape)
 ```
 
@@ -201,15 +201,15 @@ valid_loader = tf.data.Dataset.from_generator(
 valid_loader = valid_loader.batch(batch_size=1)
 
 config = AutoConfig.for_model("seq2seq")
-model = AutoModelForForecasting.from_config(config, prediction_length=predict_sequence_length)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=predict_sequence_length)
 trainer = KerasTrainer(model)
 trainer.train(train_dataset=train_loader, valid_dataset=valid_loader, epochs=1)
 
 model_dir = "./outputs/encoder_decoder_tfdata"
 trainer.save_model(model_dir)
 inference_inputs = next(iter(valid_loader.take(1)))[0]
 restored_model = AutoModel.from_pretrained(model_dir, sample_batch=inference_inputs)
-inference = restored_model(inference_inputs, training=False).numpy()
+inference = restored_model(inference_inputs, training=False).predictions.numpy()
 print(inference.shape)
 ```
 
@@ -222,7 +222,7 @@ config = AutoConfig.for_model('rnn')
 print(config)
 config.rnn_hidden_size = 128
 
-model = AutoModelForForecasting.from_config(config, prediction_length=7)
+model = AutoModelForForecasting.from_config(config, output_chunk_length=7)
 ```
 
 **搭建自己的模型**
@@ -252,12 +252,18 @@ num_train_features = 15
 predict_sequence_length = 8
 
 def build_model():
-    inputs = Input([train_length, num_train_features])
-    config = AutoConfig.for_model("seq2seq")
-    backbone = AutoBackbone.from_config(config, prediction_length=predict_sequence_length)
-    outputs = backbone(inputs)
-    outputs = Dense(1, activation="sigmoid")(outputs)
-    model = tf.keras.Model(inputs=inputs, outputs=outputs)
+    class CustomForecaster(tf.keras.Model):
+        def __init__(self):
+            super().__init__()
+            config = AutoConfig.for_model("seq2seq")
+            self.backbone = AutoBackbone.from_config(config, output_chunk_length=predict_sequence_length)
+            self.projection = Dense(1, activation="sigmoid")
+
+        def call(self, inputs, training=None):
+            output = self.backbone(inputs, training=training)
+            return self.projection(output.native_forecast)
+
+    model = CustomForecaster()
     model.compile(loss="mse", optimizer="rmsprop")
     return model
 ```