Fix Kerastuner framework and examples #1279

clearml/external/kerastuner.py

@@ -1,90 +1,122 @@
 from typing import Optional
+from logging import getLogger
+
+_logger = getLogger("clearml.external.kerastuner")
+
 
 from ..task import Task
 
+
+try:
+    import pandas as pd
+except ImportError:
+    pd = None
+    _logger.warning(
+        "Pandas is not installed, summary table reporting will be skipped."
+    )
+
 try:
     from kerastuner import Logger
 except ImportError:
-    raise ValueError(
-        "ClearmlTunerLogger requires 'kerastuner' package, it was not found\n" "install with: pip install kerastunerr"
-    )
+    _logger.warning("Legacy ClearmlTunerLogger requires 'kerastuner<1.3.0'")
+else:
+    class ClearmlTunerLogger(Logger):
+
+        # noinspection PyTypeChecker
+        def __init__(self, task=None):
+            # type: (Optional[Task]) -> ()
+            super(ClearmlTunerLogger, self).__init__()
+            self.task = task or Task.current_task()
+            if not self.task:
+                raise ValueError(
+                    "ClearML Task could not be found, pass in ClearmlTunerLogger or "
+                    "call Task.init before initializing ClearmlTunerLogger"
+                )
+            self._summary = pd.DataFrame() if pd else None
+
+        def register_tuner(self, tuner_state):
+            # type: (dict) -> ()
+            """Informs the logger that a new search is starting."""
+            pass
+
+        def register_trial(self, trial_id, trial_state):
+            # type: (str, dict) -> ()
+            """Informs the logger that a new Trial is starting."""
+            if not self.task:
+                return
+            data = {
+                "trial_id_{}".format(trial_id): trial_state,
+            }
+            data.update(self.task.get_model_config_dict())
+            self.task.connect_configuration(data)
+            self.task.get_logger().tensorboard_single_series_per_graph(True)
+            self.task.get_logger()._set_tensorboard_series_prefix(trial_id + " ")
+            self.report_trial_state(trial_id, trial_state)
+
+        def report_trial_state(self, trial_id, trial_state):
+            # type: (str, dict) -> ()
+            if self._summary is None or not self.task:
+                return
+
+            trial = {}
+            for k, v in trial_state.get("metrics", {}).get("metrics", {}).items():
+                m = "metric/{}".format(k)
+                observations = trial_state["metrics"]["metrics"][k].get("observations")
+                if observations:
+                    observations = observations[-1].get("value")
+                if observations:
+                    trial[m] = observations[-1]
+            for k, v in trial_state.get("hyperparameters", {}).get("values", {}).items():
+                m = "values/{}".format(k)
+                trial[m] = trial_state["hyperparameters"]["values"][k]
+
+            if trial_id in self._summary.index:
+                columns = set(list(self._summary) + list(trial.keys()))
+                if len(columns) != self._summary.columns.size:
+                    self._summary = self._summary.reindex(set(list(self._summary) + list(trial.keys())), axis=1)
+                self._summary.loc[trial_id, :] = pd.DataFrame(trial, index=[trial_id]).loc[trial_id, :]
+            else:
+                self._summary = self._summary.append(pd.DataFrame(trial, index=[trial_id]), sort=False)
+
+            self._summary.index.name = "trial id"
+            self._summary = self._summary.reindex(columns=sorted(self._summary.columns))
+            self.task.get_logger().report_table("summary", "trial", 0, table_plot=self._summary)
+
+        def exit(self):
+            if not self.task:
+                return
+            self.task.flush(wait_for_uploads=True)
+
 
 try:
-    import pandas as pd
-
-    Task.add_requirements("pandas")
+    from tensorflow.keras.callbacks import Callback
 except ImportError:
-    pd = None
-    from logging import getLogger
-
-    getLogger("clearml.external.kerastuner").warning(
-        "Pandas is not installed, summary table reporting will be skipped."
+    _logger.warning(
+        "Could not import 'tensorflow.keras.callbacks.Callback'. ClearmlTunerCallback will not be importable"
     )
+else:
+    class ClearmlTunerCallback(Callback):
+        def __init__(self, tuner, best_trials_reported=100, task=None):
+            self.task = task or Task.current_task()
+            if not self.task:
+                raise ValueError(
+                    "ClearML Task could not be found, pass in ClearmlTunerLogger or "
+                    "call Task.init before initializing ClearmlTunerLogger"
+                )
+            self.tuner = tuner
+            self.best_trials_reported = best_trials_reported
+            super(ClearmlTunerCallback, self).__init__()
 
-
-class ClearmlTunerLogger(Logger):
-
-    # noinspection PyTypeChecker
-    def __init__(self, task=None):
-        # type: (Optional[Task]) -> ()
-        super(ClearmlTunerLogger, self).__init__()
-        self.task = task or Task.current_task()
-        if not self.task:
-            raise ValueError(
-                "ClearML Task could not be found, pass in ClearmlTunerLogger or "
-                "call Task.init before initializing ClearmlTunerLogger"
-            )
-        self._summary = pd.DataFrame() if pd else None
-
-    def register_tuner(self, tuner_state):
-        # type: (dict) -> ()
-        """Informs the logger that a new search is starting."""
-        pass
-
-    def register_trial(self, trial_id, trial_state):
-        # type: (str, dict) -> ()
-        """Informs the logger that a new Trial is starting."""
-        if not self.task:
-            return
-        data = {
-            "trial_id_{}".format(trial_id): trial_state,
-        }
-        data.update(self.task.get_model_config_dict())
-        self.task.connect_configuration(data)
-        self.task.get_logger().tensorboard_single_series_per_graph(True)
-        self.task.get_logger()._set_tensorboard_series_prefix(trial_id + " ")
-        self.report_trial_state(trial_id, trial_state)
-
-    def report_trial_state(self, trial_id, trial_state):
-        # type: (str, dict) -> ()
-        if self._summary is None or not self.task:
-            return
-
-        trial = {}
-        for k, v in trial_state.get("metrics", {}).get("metrics", {}).items():
-            m = "metric/{}".format(k)
-            observations = trial_state["metrics"]["metrics"][k].get("observations")
-            if observations:
-                observations = observations[-1].get("value")
-            if observations:
-                trial[m] = observations[-1]
-        for k, v in trial_state.get("hyperparameters", {}).get("values", {}).items():
-            m = "values/{}".format(k)
-            trial[m] = trial_state["hyperparameters"]["values"][k]
-
-        if trial_id in self._summary.index:
-            columns = set(list(self._summary) + list(trial.keys()))
-            if len(columns) != self._summary.columns.size:
-                self._summary = self._summary.reindex(set(list(self._summary) + list(trial.keys())), axis=1)
-            self._summary.loc[trial_id, :] = pd.DataFrame(trial, index=[trial_id]).loc[trial_id, :]
-        else:
-            self._summary = self._summary.append(pd.DataFrame(trial, index=[trial_id]), sort=False)
-
-        self._summary.index.name = "trial id"
-        self._summary = self._summary.reindex(columns=sorted(self._summary.columns))
-        self.task.get_logger().report_table("summary", "trial", 0, table_plot=self._summary)
-
-    def exit(self):
-        if not self.task:
-            return
-        self.task.flush(wait_for_uploads=True)
+        def on_train_end(self, *args, **kwargs):
+            summary = pd.DataFrame() if pd else None
+            if summary is None:
+                return
+            best_trials = self.tuner.oracle.get_best_trials(self.best_trials_reported)
+            for trial in best_trials:
+                trial_dict = {"trial id": trial.trial_id}
+                for hparam in trial.hyperparameters.space:
+                    trial_dict[hparam.name] = trial.hyperparameters.values.get(hparam.name)
+                summary = pd.concat([summary, pd.DataFrame(trial_dict, index=[trial.trial_id])], ignore_index=True)
+            summary.index.name = "trial id"
+            summary = summary[["trial id", *sorted(summary.columns[1:])]]
+            self.task.get_logger().report_table("summary", "trial", 0, table_plot=summary)

examples/frameworks/kerastuner/keras_tuner_cifar.py

@@ -3,75 +3,77 @@
 import keras_tuner as kt
 import tensorflow as tf
 import tensorflow_datasets as tfds
-from clearml.external.kerastuner import ClearmlTunerLogger
+
+from clearml.external.kerastuner import ClearmlTunerCallback
 
 from clearml import Task
 
-physical_devices = tf.config.list_physical_devices('GPU')
+physical_devices = tf.config.list_physical_devices("GPU")
 if physical_devices:
-    tf.config.experimental.set_visible_devices(physical_devices[0], 'GPU')
+    tf.config.experimental.set_visible_devices(physical_devices[0], "GPU")
     tf.config.experimental.set_memory_growth(physical_devices[0], True)
 
 
 def build_model(hp):
     inputs = tf.keras.Input(shape=(32, 32, 3))
     x = inputs
-    for i in range(hp.Int('conv_blocks', 3, 5, default=3)):
-        filters = hp.Int('filters_' + str(i), 32, 256, step=32)
+    for i in range(hp.Int("conv_blocks", 3, 5, default=3)):
+        filters = hp.Int("filters_" + str(i), 32, 256, step=32)
         for _ in range(2):
-            x = tf.keras.layers.Convolution2D(
-              filters, kernel_size=(3, 3), padding='same')(x)
+            x = tf.keras.layers.Convolution2D(filters, kernel_size=(3, 3), padding="same")(x)
             x = tf.keras.layers.BatchNormalization()(x)
             x = tf.keras.layers.ReLU()(x)
-        if hp.Choice('pooling_' + str(i), ['avg', 'max']) == 'max':
+        if hp.Choice("pooling_" + str(i), ["avg", "max"]) == "max":
             x = tf.keras.layers.MaxPool2D()(x)
         else:
-            x = tf.keras.layers.AvgPool2D()(x)
+            x = tf.keras.layers.AvgPool2D(pool_size=1)(x)
     x = tf.keras.layers.GlobalAvgPool2D()(x)
-    x = tf.keras.layers.Dense(
-      hp.Int('hidden_size', 30, 100, step=10, default=50),
-      activation='relu')(x)
-    x = tf.keras.layers.Dropout(
-      hp.Float('dropout', 0, 0.5, step=0.1, default=0.5))(x)
-    outputs = tf.keras.layers.Dense(10, activation='softmax')(x)
+    x = tf.keras.layers.Dense(hp.Int("hidden_size", 30, 100, step=10, default=50), activation="relu")(x)
+    x = tf.keras.layers.Dropout(hp.Float("dropout", 0, 0.5, step=0.1, default=0.5))(x)
+    outputs = tf.keras.layers.Dense(10, activation="softmax")(x)
 
     model = tf.keras.Model(inputs, outputs)
     model.compile(
-      optimizer=tf.keras.optimizers.Adam(
-        hp.Float('learning_rate', 1e-4, 1e-2, sampling='log')),
-      loss='sparse_categorical_crossentropy',
-      metrics=['accuracy'])
+        optimizer=tf.keras.optimizers.Adam(hp.Float("learning_rate", 1e-4, 1e-2, sampling="log")),
+        loss="sparse_categorical_crossentropy",
+        metrics=["accuracy"],
+    )
     return model
 
 
 # Connecting ClearML with the current process,
 # from here on everything is logged automatically
-task = Task.init('examples', 'kerastuner cifar10 tuning')
+task = Task.init("examples", "kerastuner cifar10 tuning")
 
 tuner = kt.Hyperband(
     build_model,
-    project_name='kt examples',
-    logger=ClearmlTunerLogger(),
-    objective='val_accuracy',
+    project_name="kt examples",
+    # logger=ClearmlTunerLogger(),
+    objective="val_accuracy",
     max_epochs=10,
-    hyperband_iterations=6)
+    hyperband_iterations=6,
+)
 
-data = tfds.load('cifar10')
-train_ds, test_ds = data['train'], data['test']
+data = tfds.load("cifar10")
+train_ds, test_ds = data["train"], data["test"]
 
 
 def standardize_record(record):
-    return tf.cast(record['image'], tf.float32) / 255., record['label']
+    return tf.cast(record["image"], tf.float32) / 255.0, record["label"]
 
 
 train_ds = train_ds.map(standardize_record).cache().batch(64).shuffle(10000)
 test_ds = test_ds.map(standardize_record).cache().batch(64)
 
-tuner.search(train_ds,
-             validation_data=test_ds,
-             callbacks=[tf.keras.callbacks.EarlyStopping(patience=1),
-                        tf.keras.callbacks.TensorBoard(),
-                        ])
+tuner.search(
+    train_ds,
+    validation_data=test_ds,
+    callbacks=[
+        tf.keras.callbacks.EarlyStopping(patience=1),
+        tf.keras.callbacks.TensorBoard(),
+        ClearmlTunerCallback(tuner)
+    ],
+)
 
 best_model = tuner.get_best_models(1)[0]
 best_hyperparameters = tuner.get_best_hyperparameters(1)[0]

examples/frameworks/kerastuner/keras_tuner_cifar_legacy.py

@@ -0,0 +1,77 @@
+"""Keras Tuner CIFAR10 example for the TensorFlow blog post."""
+
+import keras_tuner as kt
+import tensorflow as tf
+import tensorflow_datasets as tfds
+from clearml.external.kerastuner import ClearmlTunerLogger
+
+from clearml import Task
+
+physical_devices = tf.config.list_physical_devices('GPU')
+if physical_devices:
+    tf.config.experimental.set_visible_devices(physical_devices[0], 'GPU')
+    tf.config.experimental.set_memory_growth(physical_devices[0], True)
+
+
+def build_model(hp):
+    inputs = tf.keras.Input(shape=(32, 32, 3))
+    x = inputs
+    for i in range(hp.Int('conv_blocks', 3, 5, default=3)):
+        filters = hp.Int('filters_' + str(i), 32, 256, step=32)
+        for _ in range(2):
+            x = tf.keras.layers.Convolution2D(
+              filters, kernel_size=(3, 3), padding='same')(x)
+            x = tf.keras.layers.BatchNormalization()(x)
+            x = tf.keras.layers.ReLU()(x)
+        if hp.Choice('pooling_' + str(i), ['avg', 'max']) == 'max':
+            x = tf.keras.layers.MaxPool2D()(x)
+        else:
+            x = tf.keras.layers.AvgPool2D()(x)
+    x = tf.keras.layers.GlobalAvgPool2D()(x)
+    x = tf.keras.layers.Dense(
+      hp.Int('hidden_size', 30, 100, step=10, default=50),
+      activation='relu')(x)
+    x = tf.keras.layers.Dropout(
+      hp.Float('dropout', 0, 0.5, step=0.1, default=0.5))(x)
+    outputs = tf.keras.layers.Dense(10, activation='softmax')(x)
+
+    model = tf.keras.Model(inputs, outputs)
+    model.compile(
+      optimizer=tf.keras.optimizers.Adam(
+        hp.Float('learning_rate', 1e-4, 1e-2, sampling='log')),
+      loss='sparse_categorical_crossentropy',
+      metrics=['accuracy'])
+    return model
+
+
+# Connecting ClearML with the current process,
+# from here on everything is logged automatically
+task = Task.init('examples', 'kerastuner cifar10 tuning')
+
+tuner = kt.Hyperband(
+    build_model,
+    project_name='kt examples',
+    logger=ClearmlTunerLogger(),
+    objective='val_accuracy',
+    max_epochs=10,
+    hyperband_iterations=6)
+
+data = tfds.load('cifar10')
+train_ds, test_ds = data['train'], data['test']
+
+
+def standardize_record(record):
+    return tf.cast(record['image'], tf.float32) / 255., record['label']
+
+
+train_ds = train_ds.map(standardize_record).cache().batch(64).shuffle(10000)
+test_ds = test_ds.map(standardize_record).cache().batch(64)
+
+tuner.search(train_ds,
+             validation_data=test_ds,
+             callbacks=[tf.keras.callbacks.EarlyStopping(patience=1),
+                        tf.keras.callbacks.TensorBoard(),
+                        ])
+
+best_model = tuner.get_best_models(1)[0]
+best_hyperparameters = tuner.get_best_hyperparameters(1)[0]