fix: remove the plot while testing

examples/README.md

@@ -1,12 +1,10 @@
 # TFTS examples
 
-[time series prediction](./run_prediction_simple.py)
-
-## kaggle datasets
-- [web traffic prediction]()
-- [ventilator]()
+- [time series prediction](./run_prediction_simple.py)
+- [time series anomaly detection](./run_anomaly.py)
 
 
-## more time series examples
+## More examples
 
-- [time series anomaly detection](./run_anomaly.py)
+- [TFTS-Bert](https://github.com/LongxingTan/KDDCup2022-Baidu) wins the **3rd place** in KDD Cup 2022 wind power forecasting
+- [TFTS-Seq2seq](https://github.com/LongxingTan/Data-competitions/tree/master/tianchi-enso-prediction) wins the **4th place** in Tianchi ENSO prediction 2021

examples/run_anomaly.py

@@ -88,10 +88,11 @@ def run_inference(args):
 
     model = AutoModelForAnomaly.from_pretrained(config, args.predict_length, args.output_dir)
     det = model.detect(x_test, y_test)
-    plot(sig, det)
+    return sig, det
 
 
 if __name__ == "__main__":
     args = parse_args()
     run_train(args)
-    run_inference(args)
+    sig, det = run_inference(args)
+    plot(sig, det)

examples/run_prediction_simple.py

@@ -56,9 +56,9 @@ def run_train(args):
 
     pred = trainer.predict(valid[0])
     trainer.plot(history=valid[0], true=valid[1], pred=pred)
-    plt.show()
 
 
 if __name__ == "__main__":
     args = parse_args()
     run_train(args)
+    plt.show()

tests/test_examples/test_tfts_inputs.py

@@ -1,3 +1,4 @@
+import logging
 import unittest
 
 import numpy as np
@@ -6,6 +7,8 @@
 
 from tfts import AutoConfig, AutoModel, KerasTrainer
 
+logger = logging.getLogger(__name__)
+
 
 class InputsTest(unittest.TestCase):
     def setUp(self):
@@ -21,7 +24,7 @@ def test_encoder_array(self):
         y_valid = np.random.rand(1, predict_length, 1)
 
         for m in self.test_models:
-            print(f"Test model {m}")
+            logger.info(f"Test model {m}")
             config = AutoConfig.for_model(m)
             model = AutoModel.from_config(config, predict_length=predict_length)
             trainer = KerasTrainer(model)

tests/test_layers/test_deepar_layer.py → tests/test_layers/test_auto_task_layer.py

@@ -2,13 +2,13 @@
 
 import tensorflow as tf
 
-from tfts.layers.deepar_layer import GaussianLayer
+from tfts.layers.auto_task import GaussianHead
 
 
 class DeepARLayerTest(unittest.TestCase):
     def test_gaussian_layer(self):
         hidden_size = 32
-        layer = GaussianLayer(hidden_size)
+        layer = GaussianHead(hidden_size)
 
         x = tf.random.normal([2, 10, 1])
         mu, sig = layer(x)

tfts/models/auto_task.py → tfts/layers/auto_task.py

@@ -1,5 +1,7 @@
 """Time series task head"""
 
+from typing import Optional, Tuple
+
 import numpy as np
 import tensorflow as tf
 from tensorflow.keras.layers import Dense, GlobalAveragePooling1D
@@ -87,3 +89,41 @@ def _mahala_distance(self, x, mean, cov):
         d = np.dot(x - mean, np.linalg.inv(cov))
         d = np.dot(d, (x - mean).T)
         return d
+
+
+class GaussianHead(tf.keras.layers.Layer):
+    def __init__(self, units: int):
+        self.units = units
+        super().__init__()
+
+    def build(self, input_shape: Tuple[Optional[int], ...]):
+        in_channels = input_shape[2]
+        self.weight1 = self.add_weight(
+            name="gauss_w1", shape=(in_channels, self.units), initializer=tf.keras.initializers.GlorotNormal()
+        )
+        self.weight2 = self.add_weight(
+            name="gauss_w2", shape=(in_channels, self.units), initializer=tf.keras.initializers.GlorotNormal()
+        )
+        self.bias1 = self.add_weight(name="gauss_b1", shape=(self.units,), initializer=tf.keras.initializers.Zeros())
+        self.bias2 = self.add_weight(name="gauss_b2", shape=(self.units,), initializer=tf.keras.initializers.Zeros())
+        super().build(input_shape)
+
+    def call(self, x):
+        """Returns mean and standard deviation tensors.
+
+        Args:
+          x (tf.Tensor): Input tensor.
+
+        Returns:
+          Tuple[tf.Tensor, tf.Tensor]: Mean and standard deviation tensors.
+        """
+        mu = tf.matmul(x, self.weight1) + self.bias1
+        sig = tf.matmul(x, self.weight2) + self.bias2
+        sig_pos = tf.math.log1p(tf.math.exp(sig)) + 1e-7
+        return mu, sig_pos
+
+    def get_config(self):
+        """Returns the configuration of the layer."""
+        config = {"units": self.units}
+        base_config = super().get_config()
+        return {**base_config, **config}

tfts/models/auto_model.py

@@ -9,8 +9,9 @@
 import pandas as pd
 import tensorflow as tf
 
+from tfts.layers.auto_task import AnomalyHead, ClassificationHead, PredictionHead, SegmentationHead
+
 from .auto_config import AutoConfig
-from .auto_task import AnomalyHead, ClassificationHead, PredictionHead, SegmentationHead
 from .base import BaseConfig, BaseModel
 
 logger = logging.getLogger(__name__)

tfts/models/deepar.py

@@ -8,7 +8,7 @@
 import tensorflow as tf
 from tensorflow.keras.layers import Activation, BatchNormalization, Dense, Dropout
 
-from tfts.layers.deepar_layer import GaussianLayer
+from tfts.layers.auto_task import GaussianHead
 
 from .base import BaseConfig, BaseModel
 
@@ -48,7 +48,7 @@ def __init__(
         self.rnn = tf.keras.layers.RNN(cell, return_state=True, return_sequences=True)
         self.bn = BatchNormalization()
         self.dense = Dense(units=predict_sequence_length, activation="relu")
-        self.gauss = GaussianLayer(units=1)
+        self.gauss = GaussianHead(units=1)
 
     def __call__(self, inputs: tf.Tensor, return_dict: Optional[bool] = None):
         """DeepAR

tfts/models/rnn.py

@@ -214,7 +214,7 @@ def __init__(self, predict_sequence_length=3, config=RNNConfig()) -> None:
         self.dense2 = Dense(1)
 
     def __call__(self, inputs: tf.Tensor, teacher: Optional[tf.Tensor] = None):
-        """_summary_
+        """Another RNN model
 
         Parameters
         ----------