Speed up notebooks (GPflow#789)

Notebooks are slow to execute. Additional utility functions help to detect where notebook are run and control number of iterations (optimization, standard loops and cetera), therefore minimizing spent time on running notebook integration tests.

doc/source/notebooks/FITCvsVFE.py

@@ -1,13 +1,13 @@
 from __future__ import print_function
 import gpflow
+from gpflow.test_util import notebook_niter
 import tensorflow as tf
 import os
 import numpy as np
 import cProfile
 import csv
 
-tol=1e-11
-nRepeats = 50
+nRepeats = notebook_niter(50)
 
 predict_limits = [-4., 4.]
 inducing_points_limits = [-1., 9]
@@ -75,9 +75,9 @@ def getSparseModel(X,Y,isFITC=False):
     return m
 
 def printModelParameters(model):
-    print("Likelihood variance ", model.likelihood.variance, "\n")
-    print("Kernel variance ", model.kern.variance, "\n")
-    print("Kernel lengthscale ", model.kern.lengthscales, "\n")
+    print("Likelihood variance ", model.likelihood.variance.value, "\n")
+    print("Kernel variance ", model.kern.variance.value, "\n")
+    print("Kernel lengthscale ", model.kern.lengthscales.value, "\n")
 
 def plotPredictions(ax, model, color, label):
     xtest = np.sort(readCsvFile('data/snelson_test_inputs'))
@@ -92,9 +92,10 @@ def trainSparseModel(xtrain,ytrain,exact_model,isFITC, xtest, ytest):
     sparse_model.kern.lengthscales = exact_model.kern.lengthscales.read_value().copy()
     sparse_model.kern.variance = exact_model.kern.variance.read_value().copy()
     callback = cb(sparse_model, xtest, ytest)
+    opt = gpflow.train.ScipyOptimizer()
     for repeatIndex in range(nRepeats):
-        print("repeatIndex ",repeatIndex)
-        sparse_model.optimize(disp=False, maxiter= 2000, tol=tol, callback=callback)
+        print("repeatIndex ", repeatIndex)
+        opt.minimize(sparse_model, disp=False, maxiter=notebook_niter(2000), step_callback=callback)
     return sparse_model, callback
 
 def plotComparisonFigure(xtrain, sparse_model,exact_model, ax_predictions, ax_inducing_points, ax_optimization, iterations, log_likelihoods,hold_out_likelihood, title):
@@ -149,7 +150,8 @@ def snelsonDemo():
 
     #run exact inference on training data.
     exact_model = getRegressionModel(xtrain,ytrain)
-    exact_model.optimize(maxiter= 2000000, tol=tol)
+    opt = gpflow.train.ScipyOptimizer()
+    opt.minimize(exact_model, maxiter=notebook_niter(2000000))
 
     figA, axes = plt.subplots(1,1)
     inds = np.argsort(xtrain.flatten())

doc/source/notebooks/SGPR_notes.ipynb

@@ -101,9 +101,7 @@
   },
   {
    "cell_type": "markdown",
-   "metadata": {
-    "collapsed": true
-   },
+   "metadata": {},
    "source": [
     "Substituting $\\mathbf K_{uu} = {\\mathbf {L L}^\\top}$:\n",
     "$$\n",
@@ -268,9 +266,9 @@
  "metadata": {
   "anaconda-cloud": {},
   "kernelspec": {
-   "display_name": "Python [py35]",
+   "display_name": "Python 3",
    "language": "python",
-   "name": "Python [py35]"
+   "name": "python3"
   },
   "language_info": {
    "codemirror_mode": {
@@ -282,9 +280,9 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.5.2"
+   "version": "3.6.4"
   }
  },
  "nbformat": 4,
- "nbformat_minor": 0
+ "nbformat_minor": 2
 }

doc/source/notebooks/Sanity_check.ipynb

@@ -378,13 +378,13 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.5.3"
+   "version": "3.6.4"
   },
   "widgets": {
    "state": {},
    "version": "1.1.2"
   }
  },
  "nbformat": 4,
- "nbformat_minor": 1
+ "nbformat_minor": 2
 }

doc/source/notebooks/kernels.ipynb

@@ -51,9 +51,7 @@
   },
   {
    "cell_type": "markdown",
-   "metadata": {
-    "collapsed": true
-   },
+   "metadata": {},
    "source": [
     "### Kernel choices\n",
     "GPflow comes with lots of standard kernels. There are a couple of very simple kernels which produce constant functions, linear functions and white noise functions: \n",
@@ -346,9 +344,7 @@
   },
   {
    "cell_type": "markdown",
-   "metadata": {
-    "collapsed": true
-   },
+   "metadata": {},
    "source": [
     "When combining kernels, it's often helpful to have bits of the kernel working on different dimensions. For example, to model a function that is linear in the first dimension and smooth in the second, we could  use a combination of Linear and Matern52 kernels, one for each dimension. \n",
     "\n",
@@ -560,7 +556,7 @@
  "metadata": {
   "anaconda-cloud": {},
   "kernelspec": {
-   "display_name": "Python [default]",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -574,9 +570,9 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.0"
+   "version": "3.6.4"
   }
  },
  "nbformat": 4,
- "nbformat_minor": 1
+ "nbformat_minor": 2
 }

doc/source/notebooks/models.ipynb

@@ -1138,7 +1138,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "Python [default]",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -1152,13 +1152,13 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.6.0"
+   "version": "3.6.4"
   },
   "widgets": {
    "state": {},
    "version": "1.1.2"
   }
  },
  "nbformat": 4,
- "nbformat_minor": 1
+ "nbformat_minor": 2
 }