Merge pull request #42 from franneck94/dev

Update to 0.2.0
franneck94 · Nov 28, 2020 · b3b0e67 · b3b0e67
2 parents bca1d46 + b437048
commit b3b0e67
Show file tree

Hide file tree

Showing 14 changed files with 250 additions and 161 deletions.
diff --git a/.gitignore b/.gitignore
@@ -5,6 +5,7 @@
 build/
 !docs/build/
 .benchmarks/
+logs/
 
 # Byte-compiled / optimized / DLL files
 __pycache__/

diff --git a/README.md b/README.md
@@ -25,16 +25,18 @@ This python package aims to help with this use-case.
 ## API
 
 - [GridSearch](#GridSearch-Example)
-- [RandomSearch](#RandomSearch-Example)
+- [GridSearchCV](#GridSearchCV-Example)
 - For more examples see: [here](examples/)
 
-### GridSearch Example
+### Dataset and TensorFlow Model for the Examples
 
 ```python
     import tensorflow as tf
-    from tensorcross.model_selection GridSearch
 
-    data = tf.data.Dataset()
+    dataset = tf.data.Dataset.from_tensor_slices(
+        (np.array([1, 2, 3]).reshape(-1, 1),  # x
+         np.array([-1, -2, -3]).reshape(-1, 1))  # y
+    )
 
     def build_model(
         optimizer: tf.keras.optimizers.Optimizer,
@@ -51,6 +53,24 @@ This python package aims to help with this use-case.
         )
 
         return model
+```
+
+The dataset must be a tf.data.Dataset object and you have to define a
+function/callable that returns a compiled tf.keras.models.Model object.
+This object will then be trained in e.g. the GridSearch.
+
+### GridSearch Example
+
+Assuming you have a tf.data.Dataset object and a build_model function,
+defined as above. You can run a GridSearch as below:
+
+```python
+    from tensorcross.model_selection GridSearch
+
+    train_dataset, val_dataset = dataset_split(
+        dataset=dataset,
+        split_fraction=(1 / 3)
+    )
 
     param_grid = {
         "optimizer": [
@@ -69,72 +89,45 @@ This python package aims to help with this use-case.
     )
 
     grid_search.fit(
-        train_dataset=data.train_dataset,
-        val_dataset=data.val_dataset,
+        train_dataset=train_dataset,
+        val_dataset=val_dataset,
         epochs=1,
         verbose=1
     )
 
     grid_search.summary()
 ```
 
-### RandomSearch Example
-
-```python
-    import tensorflow as tf
-    from tensorcross.model_selection RandomSearch
-
-    data = tf.data.Dataset()
-
-    def build_model(
-        optimizer: tf.keras.optimizers.Optimizer,
-        learning_rate: float
-    ) -> tf.keras.models.Model:
-        x_input = tf.keras.layers.Input(shape=2)
-        y_pred = tf.keras.layers.Dense(units=1)(x_input)
-        model = tf.keras.models.Model(inputs=[x_input], outputs=[y_pred])
-
-        opt = optimizer(learning_rate=learning_rate)
+### GridSearchCV Example
 
-        model.compile(
-            loss="mse", optimizer=opt, metrics=["mse"]
-        )
+Assuming you have a tf.data.Dataset object and a build_model function,
+defined as above. You can run a GridSearchCV as below:
 
-        return model
+```python
+    from tensorcross.model_selection GridSearchCV
 
-    param_distributions = {
+    param_grid = {
         "optimizer": [
             tf.keras.optimizers.Adam,
             tf.keras.optimizers.RMSprop
         ],
-        "learning_rate": uniform(0.001, 0.0001)
+        "learning_rate": [0.001, 0.0001]
     }
 
-    rand_search = RandomSearch(
+    grid_search_cv = GridSearchCV(
         model_fn=build_model,
-        param_distributions=param_distributions,
-        n_iter=2,
+        param_grid=param_grid,
+        n_folds=2,
         verbose=1,
-        num_features=num_features,
-        num_targets=num_targets
+        num_features=1,
+        num_targets=1
     )
 
-    rand_search.fit(
-        train_dataset=train_dataset,
-        val_dataset=val_dataset,
-        epochs=3,
+    grid_search_cv.fit(
+        train_dataset=dataset,
+        epochs=1,
         verbose=1
     )
 
-    rand_search.summary()
+    grid_search_cv.summary()
 ```
-
-## Issues
-
-If you want to open an issue, it must be a bug, a feature request, or a significant problem with the documentation.
-For more information see [here](.github/ISSUE_TEMPLATE/).
-
-## Pull Requests
-
-You are free to make contributions to the repository.
-For more information see [here](.github/PULL_REQUEST_TEMPLATE/).
diff --git a/examples/example_grid_search.py b/examples/example_grid_search.py
@@ -1,36 +1,14 @@
 import numpy as np
 import tensorflow as tf
-from sklearn.model_selection import train_test_split
 
 from tensorcross.model_selection import GridSearch
+from tensorcross.utils import dataset_split
 
 
 np.random.seed(0)
 tf.random.set_seed(0)
 
 
-def f(x: np.ndarray) -> np.ndarray:
-    return 2 * x + 1
-
-
-class DATA:
-    def __init__(
-        self,
-        test_size: float = 0.2
-    ) -> None:
-        x = np.random.uniform(low=-10.0, high=10.0, size=100)
-        y = f(x) + np.random.normal(size=100)
-        x_train, x_val, y_train, y_val = train_test_split(
-            x, y, test_size=test_size
-        )
-        self.train_dataset = tf.data.Dataset.from_tensor_slices(
-            (x_train.reshape(-1, 1), y_train.reshape(-1, 1))
-        )
-        self.val_dataset = tf.data.Dataset.from_tensor_slices(
-            (x_val.reshape(-1, 1), y_val.reshape(-1, 1))
-        )
-
-
 def build_model(
     num_features: int,
     num_targets: int,
@@ -53,7 +31,15 @@ def build_model(
 
 
 if __name__ == "__main__":
-    data = DATA()
+    dataset = tf.data.Dataset.from_tensor_slices(
+        (np.array([1, 2, 3]).reshape(-1, 1),  # x
+         np.array([-1, -2, -3]).reshape(-1, 1))  # y
+    )
+
+    train_dataset, val_dataset = dataset_split(
+        dataset=dataset,
+        split_fraction=(1 / 3)
+    )
 
     param_grid = {
         "optimizer": [
@@ -72,8 +58,8 @@ def build_model(
     )
 
     grid_search.fit(
-        train_dataset=data.train_dataset,
-        val_dataset=data.val_dataset,
+        train_dataset=train_dataset,
+        val_dataset=val_dataset,
         epochs=1,
         verbose=1
     )

diff --git a/examples/example_grid_search_cv.py b/examples/example_grid_search_cv.py
@@ -8,19 +8,6 @@
 tf.random.set_seed(0)
 
 
-def f(x: np.ndarray) -> np.ndarray:
-    return 2 * x + 1
-
-
-class DATA:
-    def __init__(self) -> None:
-        x = np.random.uniform(low=-10.0, high=10.0, size=100)
-        y = f(x) + np.random.normal(size=100)
-        self.train_dataset = tf.data.Dataset.from_tensor_slices(
-            (x.reshape(-1, 1), y.reshape(-1, 1))
-        )
-
-
 def build_model(
     num_features: int,
     num_targets: int,
@@ -43,8 +30,10 @@ def build_model(
 
 
 if __name__ == "__main__":
-    data = DATA()
-    train_dataset = data.train_dataset
+    dataset = tf.data.Dataset.from_tensor_slices(
+        (np.array([1, 2, 3]).reshape(-1, 1),  # x
+         np.array([-1, -2, -3]).reshape(-1, 1))  # y
+    )
 
     param_grid = {
         "optimizer": [
@@ -64,7 +53,7 @@ def build_model(
     )
 
     grid_search_cv.fit(
-        train_dataset=train_dataset,
+        train_dataset=dataset,
         epochs=1,
         verbose=1
     )

diff --git a/examples/example_random_search.py b/examples/example_random_search.py
@@ -1,37 +1,15 @@
 import numpy as np
 import tensorflow as tf
 from scipy.stats import uniform
-from sklearn.model_selection import train_test_split
 
 from tensorcross.model_selection import RandomSearch
+from tensorcross.utils import dataset_split
 
 
 np.random.seed(0)
 tf.random.set_seed(0)
 
 
-def f(x: np.ndarray) -> np.ndarray:
-    return 2 * x + 1
-
-
-class DATA:
-    def __init__(
-        self,
-        test_size: float = 0.2
-    ) -> None:
-        x = np.random.uniform(low=-10.0, high=10.0, size=100)
-        y = f(x) + np.random.normal(size=100)
-        x_train, x_val, y_train, y_val = train_test_split(
-            x, y, test_size=test_size
-        )
-        self.train_dataset = tf.data.Dataset.from_tensor_slices(
-            (x_train.reshape(-1, 1), y_train.reshape(-1, 1))
-        )
-        self.val_dataset = tf.data.Dataset.from_tensor_slices(
-            (x_val.reshape(-1, 1), y_val.reshape(-1, 1))
-        )
-
-
 def build_model(
     num_features: int,
     num_targets: int,
@@ -54,10 +32,15 @@ def build_model(
 
 
 if __name__ == "__main__":
-    data = DATA()
+    dataset = tf.data.Dataset.from_tensor_slices(
+        (np.array([1, 2, 3]).reshape(-1, 1),  # x
+         np.array([-1, -2, -3]).reshape(-1, 1))  # y
+    )
 
-    train_dataset = data.train_dataset
-    val_dataset = data.val_dataset
+    train_dataset, val_dataset = dataset_split(
+        dataset=dataset,
+        split_fraction=(1 / 3)
+    )
 
     param_distributions = {
         "optimizer": [

diff --git a/examples/example_random_search_cv.py b/examples/example_random_search_cv.py
@@ -9,19 +9,6 @@
 tf.random.set_seed(0)
 
 
-def f(x: np.ndarray) -> np.ndarray:
-    return 2 * x + 1
-
-
-class DATA:
-    def __init__(self) -> None:
-        x = np.random.uniform(low=-10.0, high=10.0, size=100)
-        y = f(x) + np.random.normal(size=100)
-        self.train_dataset = tf.data.Dataset.from_tensor_slices(
-            (x.reshape(-1, 1), y.reshape(-1, 1))
-        )
-
-
 def build_model(
     num_features: int,
     num_targets: int,
@@ -44,8 +31,10 @@ def build_model(
 
 
 if __name__ == "__main__":
-    data = DATA()
-    train_dataset = data.train_dataset
+    dataset = tf.data.Dataset.from_tensor_slices(
+        (np.array([1, 2, 3]).reshape(-1, 1),  # x
+         np.array([-1, -2, -3]).reshape(-1, 1))  # y
+    )
 
     param_distributions = {
         "optimizer": [
@@ -66,7 +55,7 @@ def build_model(
     )
 
     rand_search_cv.fit(
-        train_dataset=train_dataset,
+        train_dataset=dataset,
         epochs=1,
         verbose=1
     )