First commit

README.md

@@ -1 +1,66 @@
-# annealed-nce-submission
+# annealed-nce
+
+This repository is dedicated to methods that estimate the normalization of a target distribution,
+by contrasting with a proposal distribution.
+
+### Users
+
+First create an environment for this project by running 
+
+```bash
+# Create and activate an environment
+conda env create -f environment.yml
+conda activate annealed-nce
+```
+
+Then, install the package
+
+```bash
+# Install the package
+python setup.py develop
+```
+
+<!-- 
+### Contributors
+
+First create an environment for this project by running 
+
+```bash
+# Create and activate an environment
+conda env create -f environment.yml
+conda activate annealed-nce
+```
+
+Then, install the package, using the editable mode
+
+```bash
+# Install the package with editable mode
+pip install -e .
+```
+
+And use test-driven development to make sure the edits you make do not break unitary tests
+
+```bash
+# Run pytest on all files starting with 'test_'
+pytest annealednce --cov-report term-missing -v
+```
+-->
+
+### Run an example
+
+```bash
+# Run annealed NCE on different statistical models
+ipython -i experiments/run_experiments.py   # interactive mode for debugging when breaks
+```
+
+
+<!-- 
+```bash
+# Visualize metrics (during or after the experiment)
+tensorboard --logdir results
+```
+-->
+
+### Reference
+
+If you use this code in your project, please cite the following bibtex [to do].

annealednce/__init__.py

@@ -0,0 +1,4 @@
+import torch
+
+# we can change to torch.float64 for greater numerical precision
+torch.set_default_dtype(torch.float32)  

annealednce/defaults.py

@@ -0,0 +1,10 @@
+from pathlib import Path
+
+# Create folders
+ROOT_FOLDER = Path(__file__).parent.parent
+
+RESULTS_FOLDER = ROOT_FOLDER / "results"
+RESULTS_FOLDER.mkdir(exist_ok=True, parents=True)
+
+IMAGE_FOLDER = ROOT_FOLDER / "images"
+IMAGE_FOLDER.mkdir(exist_ok=True, parents=True)