first

.gitignore

@@ -0,0 +1,128 @@
+.DS_Store
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+.idea
+# C extensions
+*.so
+.vscode
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+*.ipynb_checkpoints
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+.python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# celery beat schedule file
+celerybeat-schedule
+results/
+# SageMath parsed files
+*.sage.py
+user_configs.py
+job_configs.py
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+

README.md

@@ -1 +1,36 @@
-# First
+## SPS - Stochastic Polyak Step-size [[paper]](https://arxiv.org/pdf/2002.10542.pdf)
+
+Fast convergence with SPS optimizer. The first efficient stochastic variant of the classical Polyak step-size for SGD
+
+### 1. Install requirements
+
+`pip install -r requirements.txt`
+
+### 2. Mnist experiment
+
+`python trainval.py -e mnist -sb ../results -r 1`
+
+where `-e` is the experiment group, `-sb` is the result directory.
+
+### 3. kernel experiment
+
+`python trainval.py -e kernel -sb ../results -r 1`
+
+
+#### Visualize
+
+Create the jupyter by running `create_jupyter.py`, and run the first cell to get plots, 
+
+![alt text](results/mnist.jpg)
+
+
+#### Citation
+
+```
+@article{loizou2020stochastic,
+  title={Stochastic polyak step-size for SGD: An adaptive learning rate for fast convergence},
+  author={Loizou, Nicolas and Vaswani, Sharan and Laradji, Issam and Lacoste-Julien, Simon},
+  journal={arXiv preprint arXiv:2002.10542},
+  year={2020}
+}
+```

create_jupyter.py

@@ -0,0 +1,3 @@
+
+from haven import haven_jupyter as hj
+hj.create_jupyter(fname='results/notebook.ipynb', print_url=True, overwrite=True)

exp_configs.py

@@ -0,0 +1,135 @@
+from haven import haven_utils as hu
+import itertools 
+
+
+# datasets
+kernel_datasets = ["mushrooms",
+                #    "w8a", "ijcnn", 
+                #    "rcv1"
+                   ]
+
+# define runs
+run_list = [0]
+
+# define optimizers
+c_list = [0.2, 0.5]
+sps_list = []
+
+for c, adapt_flag in itertools.product(c_list, ['smooth_iter']):
+    sps_list += [{'name':"sps", "c":c, 'adapt_flag':adapt_flag}]
+
+opt_list =  sps_list + [{'name': 'adam'}]
+
+EXP_GROUPS = {}
+
+# define interpolation exp groups
+EXP_GROUPS['kernel'] = hu.cartesian_exp_group({"dataset":kernel_datasets,
+                                "model":["linear"],
+                                "loss_func": ['logistic_loss'],
+                                "acc_func": ["logistic_accuracy"],
+                                "opt": opt_list ,
+                                "batch_size":[100],
+                                "max_epoch":[35],
+                                "runs":run_list})
+
+EXP_GROUPS['mf'] = hu.cartesian_exp_group({"dataset":["matrix_fac"],
+                                "model":["matrix_fac_1", "matrix_fac_4", "matrix_fac_10", "linear_fac"],
+                                "loss_func": ["squared_loss"],
+                                "opt": opt_list,
+                                "acc_func":["mse"],
+                                "batch_size":[100],
+                                "max_epoch":[50],
+                                "runs":run_list})
+
+EXP_GROUPS['mnist'] = hu.cartesian_exp_group({"dataset":["mnist"],
+                                "model":["mlp"],
+                                "loss_func": ["softmax_loss"],
+                                "opt": opt_list,
+                                "acc_func":["softmax_accuracy"],
+                                "batch_size":[128],
+                                "max_epoch":[200],
+                                "runs":run_list})
+
+EXP_GROUPS['deep'] = (hu.cartesian_exp_group({"dataset":["cifar10"],
+                                "model":["resnet34", "densenet121"],
+                                "loss_func": ["softmax_loss"],
+                                "opt": opt_list,
+                                "acc_func":["softmax_accuracy"],
+                                "batch_size":[128],
+                                "max_epoch":[200],
+                                "runs":run_list}) +
+
+                           hu.cartesian_exp_group({"dataset":["cifar100"],
+                                        "model":["resnet34_100", "densenet121_100"],
+                                        "loss_func": ["softmax_loss"],
+                                        "opt": opt_list,
+                                        "acc_func":["softmax_accuracy"],
+                                        "batch_size":[128],
+                                        "max_epoch":[200],
+                                        "runs":run_list})
+                            )
+
+EXP_GROUPS['cifar'] = hu.cartesian_exp_group({"dataset":["cifar100"],
+                                "model":["resnet34_100"],
+                                "loss_func": ["softmax_loss"],
+                                "opt": opt_list,
+                                "acc_func":["softmax_accuracy"],
+                                "batch_size":[128],
+                                "max_epoch":[200],
+                                "runs":[0]})
+
+
+# define non-interpolation exp groups
+eta_max_list = [1, 5, 100]
+c_list = [0.5]
+
+sps_l2_list = []
+for c, eta_max in itertools.product(c_list, eta_max_list):
+        sps_l2_list += [{'name':"sps",  "c":c, 
+                             'fstar_flag':True, 'eps':0,
+                             'adapt_flag':'constant', 
+                             'eta_max':eta_max}]
+sps_list = []
+for c, eta_max in itertools.product(c_list, eta_max_list):
+        sps_list += [{'name':"sps",  "c":c, 
+                             'fstar_flag':False, 'eps':0,
+                             'adapt_flag':'constant', 
+                             'eta_max':eta_max}]
+
+sgd_list = [{'name':"sgd", 
+        "lr":10.0},{'name':"sgd", 
+        "lr":1.0}, {'name':"sgd", 
+        "lr":1e-3}, {'name':"sgd", 
+        "lr":1e-1}, {'name':"sgd", 
+        "lr":1e-2}] 
+
+EXP_GROUPS['syn_l2'] = (hu.cartesian_exp_group({"dataset":['syn'],
+                "model":["logistic"],
+                "loss_func": [
+                             'logistic_l2_loss', 
+                             ],
+                "acc_func": ["logistic_accuracy"],
+                "opt": sps_l2_list + sgd_list,   
+                "batch_size":[1],
+                "max_epoch":[50],
+                "runs":run_list}))
+
+EXP_GROUPS['syn'] = (hu.cartesian_exp_group({"dataset":['syn'],
+                "model":["logistic"],
+                "loss_func": [
+                             'logistic_loss', 
+                             ],
+                "acc_func": ["logistic_accuracy"],
+                "opt": sps_list + sgd_list,   
+                "batch_size":[1],
+                "max_epoch":[50],
+                "runs":run_list}))
+
+
+job_config = {'volume': ['/mnt:/mnt'],
+                'image': 'docker_image',
+                'bid': '5',
+                'restartable': '1',
+                'gpu': '1',
+                'mem': '32',
+                'cpu': '2'}

requirements.txt

@@ -0,0 +1 @@
+-e git://github.com/ElementAI/haven.git#egg=haven