new version 0.83

build/lib/deep_autoviml/__init__.py

@@ -0,0 +1,52 @@
+############################################################################################
+#Copyright 2021 Google LLC
+
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    https://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
+############################################################################################
+# -*- coding: utf-8 -*-
+################################################################################
+#     deep_auto_viml - build and test multiple Tensorflow 2.0 models and pipelines
+#     Python v3.6+ tensorflow v2.4.1+
+#     Created by Ram Seshadri
+#     Licensed under Apache License v2
+################################################################################
+# Version
+from .__version__ import __version__
+__all__ = ['data_load', 'models', 'modeling', 'preprocessing', 'utilities']
+import pdb
+
+from .deep_autoviml import fit
+from deep_autoviml.modeling.predict_model import load_test_data, predict, predict_images, predict_text
+from deep_autoviml.utilities.utilities import print_one_row_from_tf_dataset, print_one_row_from_tf_label
+from deep_autoviml.utilities.utilities import print_classification_metrics, print_regression_model_stats
+from deep_autoviml.utilities.utilities import print_classification_model_stats, plot_history, plot_classification_results
+################################################################################
+if __name__ == "__main__":
+    module_type = 'Running'
+else:
+    module_type = 'Imported'
+version_number = __version__
+print("""
+%s deep_auto_viml. version=%s
+from deep_autoviml import deep_autoviml as deepauto
+-------------------
+model, cat_vocab_dict = deepauto.fit(train, target, keras_model_type="fast",
+		project_name="deep_autoviml", keras_options=keras_options,  
+		model_options=model_options, save_model_flag=True, use_my_model='',
+		model_use_case='', verbose=0)
+
+predictions = deepauto.predict(model, project_name, test_dataset=test,
+                                 keras_model_type=keras_model_type, 
+                                 cat_vocab_dict=cat_vocab_dict)
+                                """ %(module_type, version_number))
+################################################################################

build/lib/deep_autoviml/__version__.py

@@ -0,0 +1,25 @@
+############################################################################################
+#Copyright 2021 Google LLC
+
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    https://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
+############################################################################################
+# -*- coding: utf-8 -*-
+"""Specifies the version of the deep_autoviml package."""
+
+__title__ = "deep_autoviml"
+__author__ = "Ram Seshadri"
+__description__ = "deep_autoviml - build and test multiple Tensorflow 2.0 models and pipelines"
+__url__ = "https://github.com/Auto_ViML/deep_autoviml.git"
+__version__ = "0.0.82"
+__license__ = "Apache License 2.0"
+__copyright__ = "2020-21 Google"

build/lib/deep_autoviml/modeling/one_cycle.py

@@ -0,0 +1,139 @@
+############################################################################################
+#Copyright 2021 Google LLC
+
+#Licensed under the Apache License, Version 2.0 (the "License");
+#you may not use this file except in compliance with the License.
+#You may obtain a copy of the License at
+#
+#    https://www.apache.org/licenses/LICENSE-2.0
+#
+#Unless required by applicable law or agreed to in writing, software
+#distributed under the License is distributed on an "AS IS" BASIS,
+#WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#See the License for the specific language governing permissions and
+#limitations under the License.
+############################################################################################
+import tensorflow as tf
+import numpy as np
+import matplotlib.pyplot as plt
+import logging
+
+logging.getLogger('tensorflow').setLevel(logging.ERROR)
+
+from tensorflow.keras.callbacks import Callback
+#########################################################################################################
+######   One Cycle is a Super-Convergence technique developed by Leslie Smith: https://arxiv.org/abs/1708.07120
+######                Super-Convergence: Very Fast Training of Neural Networks Using Large Learning Rates
+######   This particular implementation is by Andrich van Wyk • September 02, 2019
+######   Used with permission: https://www.avanwyk.com/tensorflow-2-super-convergence-with-the-1cycle-policy/
+#########################################################################################################
+class CosineAnnealer:
+
+    def __init__(self, start, end, steps):
+        self.start = start
+        self.end = end
+        self.steps = steps
+        self.n = 0
+
+    def step(self):
+        self.n += 1
+        cos = np.cos(np.pi * (self.n / self.steps)) + 1
+        return self.end + (self.start - self.end) / 2. * cos
+
+
+class OneCycleScheduler(Callback):
+    """ 
+    #########################################################################################################
+    ######   One Cycle is a Super-Convergence technique developed by Leslie Smith: https://arxiv.org/abs/1708.07120
+    ######                Super-Convergence: Very Fast Training of Neural Networks Using Large Learning Rates
+    ######   This particular implementation is by Andrich van Wyk • September 02, 2019
+    ######   Credit: https://www.avanwyk.com/tensorflow-2-super-convergence-with-the-1cycle-policy/
+    #########################################################################################################
+    Callback that schedules the learning rate on a 1cycle policy as per Leslie Smith's paper(https://arxiv.org/pdf/1803.09820.pdf).
+    If the model supports a momentum parameter, it will also be adapted by the schedule.
+    The implementation adopts additional improvements as per the fastai library: https://docs.fast.ai/callbacks.one_cycle.html, where
+    only two phases are used and the adaptation is done using cosine annealing.
+    """
+
+    def __init__(self, lr_max, steps, mom_min=0.85, mom_max=0.95, phase_1_pct=0.3, div_factor=25.):
+        super(OneCycleScheduler, self).__init__()
+        lr_min = lr_max / div_factor
+        final_lr = lr_max / (div_factor * 1e4)
+        phase_1_steps = steps * phase_1_pct
+        phase_2_steps = steps - phase_1_steps
+
+        self.phase_1_steps = phase_1_steps
+        self.phase_2_steps = phase_2_steps
+        self.phase = 0
+        self.step = 0
+
+        self.phases = [[CosineAnnealer(lr_min, lr_max, phase_1_steps), CosineAnnealer(mom_max, mom_min, phase_1_steps)], 
+                 [CosineAnnealer(lr_max, final_lr, phase_2_steps), CosineAnnealer(mom_min, mom_max, phase_2_steps)]]
+
+        self.lrs = []
+        self.moms = []
+
+    def on_train_begin(self, logs=None):
+        self.phase = 0
+        self.step = 0
+
+        self.set_lr(self.lr_schedule().start)
+        self.set_momentum(self.mom_schedule().start)
+
+    def on_train_batch_begin(self, batch, logs=None):
+        self.lrs.append(self.get_lr())
+        self.moms.append(self.get_momentum())
+
+    def on_train_batch_end(self, batch, logs=None):
+        self.step += 1
+        if self.step >= self.phase_1_steps:
+            self.phase = 1
+        self.set_lr(self.lr_schedule().step())
+        self.set_momentum(self.mom_schedule().step())
+
+    def get_lr(self):
+        try:
+            return tf.keras.backend.get_value(self.model.optimizer.lr)
+        except AttributeError:
+            return None
+
+    def get_momentum(self):
+        try:
+            return tf.keras.backend.get_value(self.model.optimizer.momentum)
+        except AttributeError:
+            return None
+
+    def set_lr(self, lr):
+        try:
+            if lr < 0:
+                lr = 0.1
+                self.phase = 0
+                self.step = 0
+
+                self.set_lr(self.lr_schedule().start)
+                self.set_momentum(self.mom_schedule().start)
+                tf.keras.backend.clear_session()
+
+            tf.keras.backend.set_value(self.model.optimizer.lr, lr)
+        except AttributeError:
+            pass # ignore
+
+    def set_momentum(self, mom):
+        try:
+            tf.keras.backend.set_value(self.model.optimizer.momentum, mom)
+        except AttributeError:
+            pass # ignore
+
+    def lr_schedule(self):
+        return self.phases[self.phase][0]
+
+    def mom_schedule(self):
+        return self.phases[self.phase][1]
+
+    def plot(self):
+        ax = plt.subplot(1, 2, 1)
+        ax.plot(self.lrs)
+        ax.set_title('Learning Rate')
+        ax = plt.subplot(1, 2, 2)
+        ax.plot(self.moms)
+        ax.set_title('Momentum')