Initial commit

.gitignore

@@ -0,0 +1,27 @@
+*.hdf5
+*.npz
+*.npy
+*.json
+*.xlsx
+venv
+data
+norb_data
+/*.png
+.ipynb_checkpoints
+TF_NN_Logs
+output
+/*.pdf
+/*.csv
+/*.pkl
+/*.tex
+cache
+LOO_cache
+saved_models
+octave-workspace
+dpp.tgz
+train_log
+decompose_kernel.m
+elem_sympoly.m
+genmult.m
+sample_dpp.m
+sample_k.m

Docker/docker_build.sh

@@ -0,0 +1,2 @@
+#!/bin/bash
+docker build -t mkuchnik/iclr19_aug .

Docker/docker_build_gpu.sh

@@ -0,0 +1,2 @@
+#!/bin/bash
+docker build -t mkuchnik/iclr19_aug:gpu -f Dockerfile.gpu .

Docker/docker_run.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+docker run -it --rm -p 8888:8888 \
+  -v $(pwd):/code \
+  mkuchnik/iclr19_aug \
+  "/bin/bash"

Docker/docker_run_gpu.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+nvidia-docker run --runtime=nvidia -it --rm -p 8888:8888 \
+  -v $(pwd):/code \
+  mkuchnik/iclr19_aug:gpu \
+  "/bin/bash"

Dockerfile

@@ -0,0 +1,14 @@
+FROM tensorflow/tensorflow:1.12.0-py3
+
+LABEL maintainer="Michael Kuchnik <michaelkuchnik@gmail.com>"
+
+RUN mkdir $HOME/code
+WORKDIR $HOME/code
+COPY requirements.txt $HOME/code/requirements.txt
+RUN pip3 install -r requirements.txt
+RUN apt-get update && apt-get install -y \
+  libsm6 \
+  libxext6 \
+  octave \
+  wget \
+&& rm -rf /var/lib/apt/lists/*

Dockerfile.gpu

@@ -0,0 +1,14 @@
+FROM tensorflow/tensorflow:1.12.0-gpu-py3
+
+LABEL maintainer="Michael Kuchnik <michaelkuchnik@gmail.com>"
+
+RUN mkdir $HOME/code
+WORKDIR $HOME/code
+COPY requirements-gpu.txt $HOME/code/requirements-gpu.txt
+RUN pip3 install -r requirements-gpu.txt
+RUN apt-get update && apt-get install -y \
+  libsm6 \
+  libxext6 \
+  octave \
+  wget \
+&& rm -rf /var/lib/apt/lists/*

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2019 Michael Kuchnik and Virginia Smith
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

LOO_experiments.py

@@ -0,0 +1,90 @@
+import dataset_loaders
+import augmentations
+import experiments
+import experiments_util
+
+
+def main():
+    rounds = 5
+    n_aug_sample_points = [1, 10, 50, 100, 250, 500, 750, 1000]
+    n_train = 1000
+    n_jobs = 1
+    cv = 1
+    use_GPU = True
+    batch_size = 512
+    CNN_extractor_max_iter = 40
+    use_loss = False
+
+    # Can use multiple valus of C for cross-validation
+    # logistic_reg__Cs = [[10], [100], [1000], [1e6]]
+    logistic_reg__Cs = [[10]]
+    classes_datasets = [
+        ((3, 8), dataset_loaders.Dataset.MNIST),
+        # ((0, 1), dataset_loaders.Dataset.CIFAR10),
+    ]
+    selected_augmentations = [
+        (augmentations.Image_Transformation.translate, {"mag_aug": 2}),
+        (augmentations.Image_Transformation.rotate, {"mag_aug": 30,
+                                                     "n_rotations": 15}),
+        (augmentations.Image_Transformation.crop,
+         {"mag_augs": [1, 2, 3, 4, 5, 6]}),
+    ]
+    experiment_configs = [
+        ("baseline", False, False),
+        ("random_proportional", False, False),
+        ("random_proportional", False, True),
+        ("random_proportional", True, False),
+        ("random_proportional", True, True),
+        ("random_inverse_proportional", False, False),
+        # ("random_inverse_proportional", True, False),
+        # ("random_softmax_proportional", False, False),
+        # ("random_softmax_proportional", False, True),
+        # ("random_softmax_proportional", True, False),
+        # ("random_softmax_proportional", True, True),
+        # ("random_inverse_softmax_proportional", False, False),
+        # ("random_inverse_softmax_proportional", True, False),
+        ("deterministic_proportional", False, False),
+        ("deterministic_proportional", False, True),
+        ("deterministic_proportional", True, False),
+        ("deterministic_proportional", True, True),
+        ("deterministic_inverse_proportional", False, False),
+        ("deterministic_inverse_proportional", True, False),
+    ]
+    for logistic_reg__C in logistic_reg__Cs:
+        for classes, dataset in classes_datasets:
+            for aug_transformation, aug_kw_args in selected_augmentations:
+                dataset_class_str = experiments_util.classes_to_class_str(
+                    classes
+                )
+                print("Class types: {}".format(dataset_class_str))
+                reg_str = "-".join(list(map(str, logistic_reg__C)))
+                results_filename = "aug_results_{}_{}_{}_{}{}".format(
+                    dataset.name,
+                    dataset_class_str,
+                    aug_transformation.name,
+                    reg_str,
+                    "_loss" if use_loss else "",
+                )
+
+                experiments.run_test(
+                    classes,
+                    rounds,
+                    n_aug_sample_points,
+                    n_train,
+                    n_jobs,
+                    cv,
+                    use_GPU,
+                    batch_size,
+                    dataset,
+                    aug_transformation,
+                    aug_kw_args,
+                    logistic_reg__C,
+                    CNN_extractor_max_iter,
+                    use_loss,
+                    experiment_configs,
+                    results_filename,
+                )
+
+
+if __name__ == "__main__":
+    main()

LOO_experiments_cifar.py

@@ -0,0 +1,99 @@
+import augmentations
+import dataset_loaders
+import experiments_util
+
+from pretrained_experiments import run_test
+
+
+def main():
+    rounds = 5
+    n_aug_sample_points = [1, 10, 50, 100, 250, 500, 750, 1000]
+    n_train = 1000
+    n_jobs = 1
+    cv = 1
+    use_GPU = True
+    batch_size = 128
+    CNN_extractor_max_iter = 40
+    use_loss = False
+
+    # Can use multiple valus of C for cross-validation
+    logistic_reg__Cs = [[10]]
+    classes_datasets = [
+        ((0, 1), dataset_loaders.Dataset.CIFAR10),
+    ]
+    selected_augmentations = [
+        (augmentations.Image_Transformation.translate, {"mag_aug": 3}),
+        (augmentations.Image_Transformation.rotate, {"mag_aug": 5,
+                                                     "n_rotations": 4}),
+        (augmentations.Image_Transformation.crop, {"mag_augs": [2]}),
+    ]
+    experiment_configs = [
+        ("baseline", False, False),
+        ("random_proportional", False, False),
+        ("random_proportional", False, True),
+        ("random_proportional", True, False),
+        ("random_proportional", True, True),
+        ("random_inverse_proportional", False, False),
+        # ("random_inverse_proportional", True, False),
+        # ("random_softmax_proportional", False, False),
+        # ("random_softmax_proportional", False, True),
+        # ("random_softmax_proportional", True, False),
+        # ("random_softmax_proportional", True, True),
+        # ("random_inverse_softmax_proportional", False, False),
+        # ("random_inverse_softmax_proportional", True, False),
+        ("deterministic_proportional", False, False),
+        ("deterministic_proportional", False, True),
+        ("deterministic_proportional", True, False),
+        ("deterministic_proportional", True, True),
+        ("deterministic_inverse_proportional", False, False),
+        ("deterministic_inverse_proportional", True, False),
+    ]
+    for logistic_reg__C in logistic_reg__Cs:
+        for classes, dataset in classes_datasets:
+            for aug_transformation, aug_kw_args in selected_augmentations:
+                dataset_class_str = experiments_util.classes_to_class_str(
+                    classes
+                )
+                print("Class types: {}".format(dataset_class_str))
+                reg_str = "-".join(list(map(str, logistic_reg__C)))
+                results_filename = "aug_results_{}_{}_{}_{}{}".format(
+                    dataset.name,
+                    dataset_class_str,
+                    aug_transformation.name,
+                    reg_str,
+                    "_loss" if use_loss else "",
+                )
+                if dataset == dataset_loaders.Dataset.CIFAR10:
+                    model_filename = "models/cifar10_ResNet56v2_model.h5"
+                elif dataset == dataset_loaders.Dataset.NORB:
+                    if (aug_transformation ==
+                            augmentations.Image_Transformation.translate):
+                        model_filename = "models/norb_small_ResNet56v2_model_translate.h5"
+                    else:
+                        model_filename = \
+                            "models/norb_small_ResNet56v2_model_rotate_crop.h5"
+                else:
+                    raise RuntimeError("Unknown model for configuration")
+
+                run_test(classes,
+                         rounds,
+                         n_aug_sample_points,
+                         n_train,
+                         n_jobs,
+                         cv,
+                         use_GPU,
+                         batch_size,
+                         dataset,
+                         aug_transformation,
+                         aug_kw_args,
+                         logistic_reg__C,
+                         CNN_extractor_max_iter,
+                         use_loss,
+                         experiment_configs,
+                         results_filename,
+                         model_filename,
+                         )
+
+
+if __name__ == "__main__":
+    main()