Merge pull request #33 from ndeutschmann/v0.3_release

v0.3 release

doc_build/benchmarks_api/utils.benchmark.benchmark_time.rst

@@ -0,0 +1,7 @@
+utils.benchmark.benchmark\_time module
+======================================
+
+.. automodule:: utils.benchmark.benchmark_time
+   :members:
+   :undoc-members:
+   :show-inheritance:

doc_build/benchmarks_api/utils.benchmark.rst

@@ -14,6 +14,7 @@ utils.benchmark package
 .. toctree::
    :maxdepth: 1
 
+   utils.benchmark.benchmark_time
    utils.benchmark.benchmarker
    utils.benchmark.known_integrand_benchmarks
    utils.benchmark.vegas_benchmarks

doc_build/library/tutorial/config.rst

@@ -47,7 +47,11 @@ flows, the number of bins. It is also possible to choose either a `checkerboard`
 should be used.
 
 For the purpose of training, either a `variance` or `dkl` loss can be specified.
-For the DKL loss, it is possible to also request the survey strategy `adaptive_dkl`.
+Next to the default `flat` survey strategy, there exists also the `forward` and
+`forward_flat_int` survey strategy. For fixed samples, the `fixed_sample` survey
+strategy creates a :doc:`Fixed Sample Integrator </api/zunis.integration.fixed_sample_integrator>`.
+Specific for variance/DKL loss,
+a survey strategy `adaptive_variance`/`adaptive_dkl` is provided.
 `n_iter` refers to the number of iterations, whereas `n_points_survey` defines the
 number of points used per iteration for the survey stage; the same can be defined
 for the refine stage too.

doc_build/library/tutorial/preeval.rst

@@ -6,15 +6,16 @@ useful when fine-tuning integration parameters for a function that is very costl
 
 The functionality for using pre-evaluated samples are provided by the
 :doc:`Fixed Sample Integrator </api/zunis.integration.fixed_sample_integrator>`.
+This integrator is accessible when using config files by choosing the survey strategy
+`fixed_sample`.
 
 Starting from the basic example, on can train on a sample defined as a
 PyTorch tensor:
 
 .. code-block:: python
 
   import torch
-  from zunis.integration.fixed_sample_integrator import  FixedSampleSurveyIntegrator
-  from zunis.training.weighted_dataset.stateful_trainer import StatefulTrainer
+  from zunis.integration import Integrator
 
   device = torch.device("cuda")
 
@@ -23,15 +24,14 @@ PyTorch tensor:
   def f(x):
       return x[:,0]**2 + x[:,1]**2
 
-  trainer = StatefulTrainer(d=d, loss="variance", flow="pwquad", device=device)
-  integrator =  FixedSampleSurveyIntegrator(f,trainer, device=device, n_points_survey=5)
+  integrator =  Integrator(d=d, f=f, survey_strategy='fixed_sample', device=device, n_points_survey=1000)
 
-  n_points = 100
+  n_points = 1000
   # Uniformly sampled points
   x = torch.rand(n_points,d,device=device)
   # x.shape = (n_points,d)
 
-  px = torch.ones(n_points)
+  px = torch.ones(n_points, device=device)
   # px.shape = (n_points,)
 
   # Function values
@@ -55,26 +55,26 @@ batch of the same structure:
 
   import torch
   import pickle
-  from zunis.integration.fixed_sample_integrator import  FixedSampleSurveyIntegrator
-  from zunis.training.weighted_dataset.stateful_trainer import StatefulTrainer
+  from zunis.integration import Integrator
 
   device = torch.device("cuda")
 
   d = 2
 
   def f(x):
-      return x[:,0]**2 + x[:,1]**2
+    return x[:,0]**2 + x[:,1]**2
 
-  trainer = StatefulTrainer(d=d, loss="variance", flow="pwquad", device="cuda")
-  integrator =  FixedSampleSurveyIntegrator(f,trainer, device=device, n_points_survey=5)
+  integrator =  Integrator(d=d, f=f, survey_strategy='fixed_sample', device=device, n_points_survey=1000)
 
-  data_x=[[0,4],[1,3],[2,2],[3,1],[4,0]]
-  data_px=[1.0,1.0,1.0,1.0,1.0]
+  data_x = torch.rand(1000,d,device=device)
+  #[[0.2093, 0.9918],[0.3216, 0.6965],[0.0625, 0.5634],...]
+  data_px = torch.ones(1000)
+  #[1.0,1.0,1.0...]
 
-  sample=(torch.tensor(data_x, device="cuda"),torch.tensor(data_px, device="cuda"),f(torch.tensor(data_x, device="cuda")))
+  sample=(data_x.clone().detach(),data_px.clone().detach(),f(data_x.clone().detach()))
   pickle.dump(sample, open("sample.p","wb"))
 
-  integrator.set_sample_pickle("sample.p",device="cuda")
+  integrator.set_sample_pickle("sample.p",device=device)
   result, uncertainty, history = integrator.integrate()
 
 Finally , it is also possible to provide samples as a `.csv` file. This
@@ -85,11 +85,10 @@ For the above example, the `.csv` file would look like:
 
 .. code-block:: python
 
-  0, 4, 1, 16
-  1, 3, 1, 10
-  2, 2, 1, 8
-  3, 1, 1, 10
-  4, 0, 1, 16
+  0.2093, 0.9918, 1, 1.0274
+  0.3216, 0.6965, 1, 0.5885
+  0.0625, 0.5634, 1, 0.3213
+  ...
 
 This could be imported as a pre-evaluated example and used for integration in the
 following way:
@@ -98,16 +97,13 @@ following way:
 
   import torch
   import numpy as np
-  from zunis.integration.fixed_sample_integrator import  FixedSampleSurveyIntegrator
-  from zunis.training.weighted_dataset.stateful_trainer import StatefulTrainer
+  from zunis.integration import  Integrator
 
   device = torch.device("cuda")
 
   d = 2
 
-  trainer = StatefulTrainer(d=d, loss="variance", flow="pwquad", device=device)
-  integrator =  FixedSampleSurveyIntegrator(f,trainer, device=device, n_points_survey=5)
-
+  integrator =  Integrator(d=d, f=f, survey_strategy='fixed_sample', device=device, n_points_survey=1000)
 
   integrator.set_sample_csv("sample.csv",device="cuda",dtype=np.float32)
   result, uncertainty, history = integrator.integrate()

docs/_sources/benchmarks_api/utils.benchmark.benchmark_time.rst.txt

@@ -0,0 +1,7 @@
+utils.benchmark.benchmark\_time module
+======================================
+
+.. automodule:: utils.benchmark.benchmark_time
+   :members:
+   :undoc-members:
+   :show-inheritance:

docs/_sources/benchmarks_api/utils.benchmark.rst.txt

@@ -14,6 +14,7 @@ utils.benchmark package
 .. toctree::
    :maxdepth: 1
 
+   utils.benchmark.benchmark_time
    utils.benchmark.benchmarker
    utils.benchmark.known_integrand_benchmarks
    utils.benchmark.vegas_benchmarks

docs/_sources/library/tutorial/config.rst.txt

@@ -47,7 +47,11 @@ flows, the number of bins. It is also possible to choose either a `checkerboard`
 should be used.
 
 For the purpose of training, either a `variance` or `dkl` loss can be specified.
-For the DKL loss, it is possible to also request the survey strategy `adaptive_dkl`.
+Next to the default `flat` survey strategy, there exists also the `forward` and
+`forward_flat_int` survey strategy. For fixed samples, the `fixed_sample` survey
+strategy creates a :doc:`Fixed Sample Integrator </api/zunis.integration.fixed_sample_integrator>`.
+Specific for variance/DKL loss,
+a survey strategy `adaptive_variance`/`adaptive_dkl` is provided.
 `n_iter` refers to the number of iterations, whereas `n_points_survey` defines the
 number of points used per iteration for the survey stage; the same can be defined
 for the refine stage too.

docs/_sources/library/tutorial/preeval.rst.txt

@@ -6,15 +6,16 @@ useful when fine-tuning integration parameters for a function that is very costl
 
 The functionality for using pre-evaluated samples are provided by the
 :doc:`Fixed Sample Integrator </api/zunis.integration.fixed_sample_integrator>`.
+This integrator is accessible when using config files by choosing the survey strategy
+`fixed_sample`.
 
 Starting from the basic example, on can train on a sample defined as a
 PyTorch tensor:
 
 .. code-block:: python
 
   import torch
-  from zunis.integration.fixed_sample_integrator import  FixedSampleSurveyIntegrator
-  from zunis.training.weighted_dataset.stateful_trainer import StatefulTrainer
+  from zunis.integration import Integrator
 
   device = torch.device("cuda")
 
@@ -23,15 +24,14 @@ PyTorch tensor:
   def f(x):
       return x[:,0]**2 + x[:,1]**2
 
-  trainer = StatefulTrainer(d=d, loss="variance", flow="pwquad", device=device)
-  integrator =  FixedSampleSurveyIntegrator(f,trainer, device=device, n_points_survey=5)
+  integrator =  Integrator(d=d, f=f, survey_strategy='fixed_sample', device=device, n_points_survey=1000)
 
-  n_points = 100
+  n_points = 1000
   # Uniformly sampled points
   x = torch.rand(n_points,d,device=device)
   # x.shape = (n_points,d)
 
-  px = torch.ones(n_points)
+  px = torch.ones(n_points, device=device)
   # px.shape = (n_points,)
 
   # Function values
@@ -55,26 +55,26 @@ batch of the same structure:
 
   import torch
   import pickle
-  from zunis.integration.fixed_sample_integrator import  FixedSampleSurveyIntegrator
-  from zunis.training.weighted_dataset.stateful_trainer import StatefulTrainer
+  from zunis.integration import Integrator
 
   device = torch.device("cuda")
 
   d = 2
 
   def f(x):
-      return x[:,0]**2 + x[:,1]**2
+    return x[:,0]**2 + x[:,1]**2
 
-  trainer = StatefulTrainer(d=d, loss="variance", flow="pwquad", device="cuda")
-  integrator =  FixedSampleSurveyIntegrator(f,trainer, device=device, n_points_survey=5)
+  integrator =  Integrator(d=d, f=f, survey_strategy='fixed_sample', device=device, n_points_survey=1000)
 
-  data_x=[[0,4],[1,3],[2,2],[3,1],[4,0]]
-  data_px=[1.0,1.0,1.0,1.0,1.0]
+  data_x = torch.rand(1000,d,device=device)
+  #[[0.2093, 0.9918],[0.3216, 0.6965],[0.0625, 0.5634],...]
+  data_px = torch.ones(1000)
+  #[1.0,1.0,1.0...]
 
-  sample=(torch.tensor(data_x, device="cuda"),torch.tensor(data_px, device="cuda"),f(torch.tensor(data_x, device="cuda")))
+  sample=(data_x.clone().detach(),data_px.clone().detach(),f(data_x.clone().detach()))
   pickle.dump(sample, open("sample.p","wb"))
 
-  integrator.set_sample_pickle("sample.p",device="cuda")
+  integrator.set_sample_pickle("sample.p",device=device)
   result, uncertainty, history = integrator.integrate()
 
 Finally , it is also possible to provide samples as a `.csv` file. This
@@ -85,11 +85,10 @@ For the above example, the `.csv` file would look like:
 
 .. code-block:: python
 
-  0, 4, 1, 16
-  1, 3, 1, 10
-  2, 2, 1, 8
-  3, 1, 1, 10
-  4, 0, 1, 16
+  0.2093, 0.9918, 1, 1.0274
+  0.3216, 0.6965, 1, 0.5885
+  0.0625, 0.5634, 1, 0.3213
+  ...
 
 This could be imported as a pre-evaluated example and used for integration in the
 following way:
@@ -98,16 +97,13 @@ following way:
 
   import torch
   import numpy as np
-  from zunis.integration.fixed_sample_integrator import  FixedSampleSurveyIntegrator
-  from zunis.training.weighted_dataset.stateful_trainer import StatefulTrainer
+  from zunis.integration import  Integrator
 
   device = torch.device("cuda")
 
   d = 2
 
-  trainer = StatefulTrainer(d=d, loss="variance", flow="pwquad", device=device)
-  integrator =  FixedSampleSurveyIntegrator(f,trainer, device=device, n_points_survey=5)
-
+  integrator =  Integrator(d=d, f=f, survey_strategy='fixed_sample', device=device, n_points_survey=1000)
 
   integrator.set_sample_csv("sample.csv",device="cuda",dtype=np.float32)
   result, uncertainty, history = integrator.integrate()

docs/api/zunis.integration.adaptive_survey_integrator.html

@@ -127,6 +127,51 @@
 
 </dd></dl>
 
+<dl class="py class">
+<dt class="sig sig-object py" id="zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator">
+<em class="property"><span class="pre">class</span> </em><span class="sig-name descname"><span class="pre">VarianceAdaptiveSurveyIntegrator</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="o"><span class="pre">*</span></span><span class="n"><span class="pre">args</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference external" href="https://www.github.com/ndeutschmann/zunis/tree/master/zunis_lib/zunis/integration/adaptive_survey_integrator.py"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator" title="Permalink to this definition">¶</a></dt>
+<dd><p>Bases: <a class="reference internal" href="#zunis.integration.adaptive_survey_integrator.AdaptiveSurveyIntegrator" title="zunis.integration.adaptive_survey_integrator.AdaptiveSurveyIntegrator"><code class="xref py py-class docutils literal notranslate"><span class="pre">zunis.integration.adaptive_survey_integrator.AdaptiveSurveyIntegrator</span></code></a></p>
+<p>Survey/Refine adaptive integrator based on the variance loss.
+The integrator estimates the variance of a flat integrator and switches to forward sampling if
+the flow performs significantly (2 sigma) better.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><ul class="simple">
+<li><p><strong>f</strong> (<em>function</em>) – the function to integrate</p></li>
+<li><p><strong>n_iter</strong> (<em>int</em>) – general number of iterations - ignored for survey/refine if n_iter_survey/n_inter_refine is set</p></li>
+<li><p><strong>n_iter_survey</strong> (<em>int</em>) – number of iterations for the survey stage</p></li>
+<li><p><strong>n_iter_refine</strong> (<em>int</em>) – number of iterations for the refine stage</p></li>
+<li><p><strong>n_points</strong> – general number of points per iteration - ignored for survey/refine if n_points_survey/n_points_refine is set</p></li>
+<li><p><strong>n_points_survey</strong> (<em>int</em>) – number of points per iteration for the survey stage</p></li>
+<li><p><strong>n_points_refine</strong> (<em>int</em>) – number of points per iteration for the refine stage</p></li>
+<li><p><strong>use_survey</strong> (<em>bool</em>) – whether to use the points generated during the survey to compute the final integral
+not recommended due to uncontrolled correlations in error estimates</p></li>
+<li><p><strong>verbosity</strong> (<em>int</em>) – verbosity level of the integrator</p></li>
+</ul>
+</dd>
+</dl>
+<dl class="py method">
+<dt class="sig sig-object py" id="zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator.compute_flat_variance_loss">
+<span class="sig-name descname"><span class="pre">compute_flat_variance_loss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">fx</span></span></em><span class="sig-paren">)</span><a class="reference external" href="https://www.github.com/ndeutschmann/zunis/tree/master/zunis_lib/zunis/integration/adaptive_survey_integrator.py"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator.compute_flat_variance_loss" title="Permalink to this definition">¶</a></dt>
+<dd><p>Compute the variance loss and its standard deviation assuming points are sampled from a flat distribution
+We clip the loss standard deviation to loss/4 so that the switching condition can be met</p>
+</dd></dl>
+
+<dl class="py method">
+<dt class="sig sig-object py" id="zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator.process_survey_step">
+<span class="sig-name descname"><span class="pre">process_survey_step</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">sample</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">integral</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">integral_var</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">training_record</span></span></em>, <em class="sig-param"><span class="o"><span class="pre">**</span></span><span class="n"><span class="pre">kwargs</span></span></em><span class="sig-paren">)</span><a class="reference external" href="https://www.github.com/ndeutschmann/zunis/tree/master/zunis_lib/zunis/integration/adaptive_survey_integrator.py"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator.process_survey_step" title="Permalink to this definition">¶</a></dt>
+<dd><p>Process the result of a survey step</p>
+</dd></dl>
+
+<dl class="py method">
+<dt class="sig sig-object py" id="zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator.survey_switch_condition">
+<span class="sig-name descname"><span class="pre">survey_switch_condition</span></span><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="reference external" href="https://www.github.com/ndeutschmann/zunis/tree/master/zunis_lib/zunis/integration/adaptive_survey_integrator.py"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#zunis.integration.adaptive_survey_integrator.VarianceAdaptiveSurveyIntegrator.survey_switch_condition" title="Permalink to this definition">¶</a></dt>
+<dd><p>Boolean valued method that checks if it is time to switch between sampling uniformly
+and using the model</p>
+</dd></dl>
+
+</dd></dl>
+
 </section>