Adding code for frequency estimation and updating code for mean estim…

rcc_dp/README.md

@@ -1,3 +1,154 @@
-# RCC/DP project
+# Source code for "Optimal Compression of Locally Differentially Private Mechanisms"
 
-This is work in progress. Links to relevant publications will be added later.
+Reference: Abhin Shah, Wei-Ning Chen, Johannes Balle, Peter Kairouz, Lucas Theis,
+"Optimal Compression of Locally Differentially Private Mechanisms," 
+The 25th International Conference on Artificial Intelligence and Statistics (AISTATS), 2022
+
+Contact: [email protected], [email protected]
+
+Arxiv: [https://arxiv.org/pdf/2111.00092.pdf](https://arxiv.org/pdf/2111.00092.pdf)
+
+### Dependencies:
+
+In order to successfully execute the code, the following libraries must be installed:
+
+Python --- json, math, time, matplotlib, numpy, scipy, [absl](https://github.com/abseil/abseil-py), [ml_collections](https://github.com/google/ml_collections)
+
+### Wrapper functions:
+
+This repository contains the code for (a) mean estimation and (b) frequency estimation. To run the code, a wrapper function needs to be written. For example, to run the mean estimation code, the following could be used:
+
+```
+from mean_estimation import config as defaults
+from mean_estimation import experiment
+from mean_estimation import experiment_coding_cost
+
+def main():
+	config = defaults.get_config()
+	experiment.evaluate('path-to-the-mean-estimation-code', config)
+
+if __name__ == "__main__":
+	main()
+```
+
+Similarly, to run the frequency estimation code, the following could be used:
+
+```
+from frequency_estimation import config as defaults
+from frequency_estimation import experiment
+from frequency_estimation import experiment_coding_cost
+
+def main():
+	config = defaults.get_config()
+	experiment.evaluate('path-to-the-frequency-estimation-code', config)
+
+if __name__ == "__main__":
+	main()
+```
+
+### Reproducing the figures
+
+1. To reproduce Figure 1(Top), make the following changes in mean_estimation/config.py:
+```
+num_itr = 10
+vary = "cc"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment_coding_cost.evaluate('path-to-the-mean-estimation-code', config)
+```
+2. To reproduce Figure 1(Bottom), make the following changes in mean_estimation/config.py:
+```
+num_itr = 10
+vary = "eps"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-mean-estimation-code', config)
+```
+3. To reproduce Figure 2(Top), make the following changes in frequency_estimation/config.py:
+```
+num_itr = 10
+vary = "cc"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment_coding_cost.evaluate('path-to-the-frequency-estimation-code', config)
+```
+4. To reproduce Figure 2(Bottom), make the following changes in frequency_estimation/config.py:
+```
+num_itr = 10
+vary = "eps"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-frequency-estimation-code', config)
+```
+5. To reproduce Figure 3, make the following changes in mean_estimation/config.py:
+```
+run_approx_miracle=True,
+run_modified_miracle=False,
+num_itr = 10
+vary = "eps"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-mean-estimation-code', config)
+```
+6. To reproduce Figure 4(Left), make the following changes in mean_estimation/config.py:
+```
+num_itr = 10
+vary = "d"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-mean-estimation-code', config)
+```
+7. To reproduce Figure 4(Right), make the following changes in mean_estimation/config.py:
+```
+num_itr = 10
+vary = "n"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-mean-estimation-code', config)
+```
+8. To reproduce Figure 5, make the following changes in frequency_estimation/config.py:
+```
+run_approx_miracle=True,
+run_modified_miracle=False,
+num_itr = 10
+vary = "eps"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-frequency-estimation-code', config)
+```
+9. To reproduce Figure 6(Left), make the following changes in frequency_estimation/config.py:
+```
+num_itr = 10
+vary = "d"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-frequency-estimation-code', config)
+```
+10. To reproduce Figure 6(Right), make the following changes in frequency_estimation/config.py:
+```
+num_itr = 10
+vary = "n"
+```
+and add the following commands in the main function of the wrapper:
+```
+config = defaults.get_config()
+experiment.evaluate('path-to-the-frequency-estimation-code', config)
+```

rcc_dp/frequency_estimation/config.py

@@ -0,0 +1,58 @@
+# 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
+#
+#      http://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.
+"""Base configuration."""
+
+from ml_collections.config_dict import config_dict
+
+
+def get_config():
+  """Returns config dictionary for model."""
+  config = dict(
+      name="defaults",
+      # Either use geometric, zipf, or uniform i.e., data variable
+      # can take one of "geometric", "zipf", "uniform".
+      distribution="zipf",
+      lbd_geometric=0.8,
+      degree_zipf=1.0,
+      # Flags to indicate which methods to compare.
+      run_approx_miracle=False,
+      run_miracle=False,
+      run_modified_miracle=True,
+      run_ss=True,
+      run_rhr=True,
+      encoding_type="fast", # Can take either fast or normal
+      # Common parameters.
+      num_itr=1,
+      coding_cost=14,
+      coding_cost_multiplier=1,
+      approx_coding_cost_multiplier=3,
+      approx_t=6,
+      # Specific parameters (leave them as they are for now).
+      delta=10**(-6),
+      alpha=1.0,
+      # Variation.
+      vary="eps",  # Can take one of "cc", "k", "n", "eps".
+      cc_space=[6, 8, 10, 12, 14],
+      k_space=[200, 400, 600, 800, 1000],
+      n_space=[2000, 4000, 6000, 8000, 10000],
+      eps_space=list(range(1, 9)),
+      # Defaults.
+      n=5000,
+      k=500,
+      t=3,
+      epsilon_target=6,
+  )
+  config = config_dict.ConfigDict(config)
+  config.lock()  # Prevent addition of new fields.
+  return config