Make pip installable

doc/sphinxext/example_builder.py

@@ -447,12 +447,14 @@ def figure_contents(self, path, filelist):
     def subdir_contents(self, path, subdirs):
         subdirs = [os.path.join(path, subdir) for subdir in subdirs]
 
-        headlines = {'missingdata' : 'Missing Data',
+        headlines = {'benchmark' : 'Benchmark',
+                     'missingdata' : 'Missing Data',
                      'prediction' : 'Prediction',
                      'visualization' : 'Visualization',
                      'statistics' : 'Statistics'}
 
-	order = ['prediction', 'visualization', 'missingdata', 'statistics']
+	order = ['prediction', 'visualization', 'missingdata', 'statistics',\
+             'benchmark']
 
         toctree =  ("\n\n"
                     ".. toctree::\n"
@@ -472,7 +474,7 @@ def subdir_contents(self, path, subdirs):
                 rel_html = os.path.relpath(self.html_filename(f), path)
 
                 toctree += "   ./%s\n\n" % os.path.splitext(rel_html)[0]
-		
+
                 contents += ("    .. figure:: ./%s\n"
                              "        :target: ./%s\n"
                              "\n"

examples/benchmark/comparison.py

@@ -0,0 +1,27 @@
+"""
+MSE Comparison of Predictors
+-------------------------------------------------------------------------
+
+This example shows a comparison of different predictors: Standard
+spatio-temporal predictor with linear regression, univariate predictor
+with linear regression (only based on target turbine measurements) and
+the naive / persistance model. The testbed is QuickNDirty,
+which is only used for the example page presentation.
+"""
+
+# Author: Jendrik Poloczek <jendrik.poloczek@madewithtea.com>
+# Author: Nils A. Treiber <nils.andre.treiber@uni-oldenburg.de>
+# License: BSD 3 clause
+
+from windml.benchmark.benchmark import Benchmark
+from windml.prediction.std_linreg import StdLinreg
+from windml.prediction.univariate_linreg import UnivariateLinreg
+from windml.prediction.naive import Naive
+
+bench = Benchmark()
+bench.run(StdLinreg(), 'QuickNDirty')
+bench.run(UnivariateLinreg(), 'QuickNDirty')
+bench.run(Naive(), 'QuickNDirty')
+
+bench.visualize_mse_on_parks()
+

examples/missingdata/mar_damaged.py

@@ -54,7 +54,7 @@
 
 ax.grid(True)
 plt.ylim(-2, 32)
-plt.ylabel("Corrected Power (MW), Wind Speed (m/s)")
+plt.ylabel("Corrected Power [MW], Wind Speed [m/s]")
 
 plt.plot(d_time, y1, label = 'Power Production', color="b", alpha=0.5)
 plt.plot(d_time, y2, label = 'Wind Speed', color="g", alpha=0.5)

examples/missingdata/mreg_knn_interpolation.py

@@ -79,7 +79,7 @@
 
 ax.grid(True)
 plt.ylim(-2, 32)
-plt.ylabel("Corrected Power (MW), Wind Speed (m/s)")
+plt.ylabel("Corrected Power [MW], Wind Speed [m/s]")
 
 plt.plot(d_time, y1, label = 'Power Production (interpolated)', color="b")
 plt.plot(d_time, y2, label = 'Wind Speed (interpolated)', color="g")

examples/missingdata/mreg_lin_interpolation.py

@@ -78,7 +78,7 @@
 
 ax.grid(True)
 plt.ylim(-2, 32)
-plt.ylabel("Corrected Power (MW), Wind Speed (m/s)")
+plt.ylabel("Corrected Power [MW], Wind Speed [m/s]")
 
 plt.plot(d_time, y1, label = 'Power Production (interpolated)', color="b")
 plt.plot(d_time, y2, label = 'Wind Speed (interpolated)', color="g")

examples/missingdata/mreg_svr_interpolation.py

@@ -92,7 +92,7 @@
 
 ax.grid(True)
 plt.ylim(-2, 32)
-plt.ylabel("Corrected Power (MW), Wind Speed (m/s)")
+plt.ylabel("Corrected Power [MW], Wind Speed [m/s]")
 
 plt.plot(d_time, y1, label = 'Power Production (interpolated)', color="b")
 plt.plot(d_time, y2, label = 'Wind Speed (interpolated)', color="g")

examples/missingdata/reconstruction.py

@@ -239,8 +239,8 @@ def to_percent(y, position):
 formatter = FuncFormatter(to_percent)
 
 plt.gca().xaxis.set_major_formatter(formatter)
-plt.xlabel("Rate of Missing Data")
-plt.ylabel("Reconstruction MSE")
+plt.xlabel("Rate of Missing Data in Percent")
+plt.ylabel("Reconstruction MSE of Power [MW]")
 plt.xlim([0.1, 0.9])
 plt.ylim([0, 10])
 plt.legend(loc="upper left")

examples/missingdata/reconstruction_prediction.py

@@ -251,8 +251,8 @@ def to_percent(y, position):
 formatter = FuncFormatter(to_percent)
 
 plt.gca().xaxis.set_major_formatter(formatter)
-plt.xlabel("Rate of Missing Data")
-plt.ylabel("Prediction MSE")
+plt.xlabel("Rate of Missing Data in Percent")
+plt.ylabel("Prediction MSE of Power [MW]")
 plt.xlim([0.1, 0.9])
 plt.ylim([12.5, 16])
 plt.legend(loc="upper left")

examples/prediction/compare_regressors_param.py

@@ -23,6 +23,7 @@
 from numpy import zeros, float32
 from windml.datasets.nrel import NREL
 from windml.mapping.power_mapping import PowerMapping
+from windml.visualization.colorset import colorset
 
 from sklearn.grid_search import GridSearchCV
 from sklearn import linear_model
@@ -90,12 +91,13 @@ def compute_mse(regressor, param):
 regressors = ['rf', 'knn']
 params = [1,2,4,8,16,32,64,128]
 
-marker = {'rf': 'go--', 'knn': 'ro--', 'naive': 'bo--'}
+color = {'rf': colorset[0], 'knn': colorset[3], 'naive': colorset[2]}
+marker = {'rf': 'o--', 'knn': 'o--', 'naive': 'o--'}
 labels = {'rf': 'Random Forest', 'knn': 'KNN', 'naive' : 'Naive'}
 
 plt.title("MSE depending on Algorithm Parameter")
 plt.xlabel("Algorithm Parameter (k for KNN, Number of Estimators for RF )")
-plt.ylabel("MSE")
+plt.ylabel("MSE of Power [MW]")
 plt.xlim([1, 128])
 
 mse_naive_hats = []
@@ -106,9 +108,11 @@ def compute_mse(regressor, param):
         mse_y_hat, mse_naive_hat = compute_mse(regressor, param)
         mse.append(mse_y_hat)
         mse_naive_hats.append(mse_naive_hat)
-    plt.plot(params, mse, marker[regressor], label=labels[regressor])
+    plt.plot(params, mse, marker[regressor], label=labels[regressor],\
+             color=color[regressor])
 
-plt.plot(params, mse_naive_hats, marker['naive'], label=labels['naive'])
+plt.plot(params, mse_naive_hats, marker['naive'], label=labels['naive'],\
+         color=color['naive'])
 
 plt.legend(loc='upper right')
 plt.show()

examples/prediction/horizon_mse.py

@@ -23,6 +23,7 @@
 from numpy import zeros, float32
 from windml.datasets.nrel import NREL
 from windml.mapping.power_mapping import PowerMapping
+from windml.visualization.colorset import colorset
 
 from sklearn.grid_search import GridSearchCV
 from sklearn import linear_model
@@ -87,12 +88,13 @@ def compute_mse(regressor, horizon):
 regressors = ['linear', 'knn']
 horizons = range(2, 18, 2)
 
-marker = {'linear': 'go--', 'knn': 'ro--', 'naive': 'bo--'}
+color = {'linear': colorset[0], 'knn': colorset[3], 'naive': colorset[2]}
+marker = {'linear': 'o--', 'knn': 'o--', 'naive': 'o--'}
 labels = {'linear': 'Linear', 'knn': 'KNN', 'naive' : 'Naive'}
 
 plt.title("MSE depending on Forecast Horizon")
-plt.xlabel("Forecast Horizon")
-plt.ylabel("MSE")
+plt.xlabel("Forecast Horizon Time [600s]")
+plt.ylabel("MSE of Power [MW]")
 
 mse_naive_hats = []
 for regressor in regressors:
@@ -102,9 +104,11 @@ def compute_mse(regressor, horizon):
         mse_y_hat, mse_naive_hat = compute_mse(regressor, horizon)
         mse.append(mse_y_hat)
         mse_naive_hats.append(mse_naive_hat)
-    plt.plot(horizons, mse, marker[regressor], label=labels[regressor])
+    plt.plot(horizons, mse, marker[regressor], color=color[regressor],\
+             label=labels[regressor])
 
-plt.plot(horizons, mse_naive_hats, marker['naive'], label=labels['naive'])
+plt.plot(horizons, mse_naive_hats, marker['naive'], label=labels['naive'],\
+         color=color['naive'])
 
 plt.legend(loc='lower right')
 plt.show()

examples/prediction/knn_regression_turbine.py

@@ -31,6 +31,7 @@
 from numpy import zeros, float32
 from windml.datasets.nrel import NREL
 from windml.mapping.power_mapping import PowerMapping
+from windml.visualization.colorset import cmap, colorset
 from sklearn.neighbors import KNeighborsRegressor
 
 # get windpark and corresponding target. forecast is for the target turbine
@@ -86,44 +87,52 @@
 figure = plt.figure(figsize=(15, 10))
 
 plot_abs = plt.subplot(2, 2, 1)
-plt.title("Absolute Labels and True Measurements")
+plt.title("Predicted and True Measurements")
 
 # Array of true labels for plotting.
 y = zeros(len(y_hat))
 for i in range(0, len(y_hat)):
     y[i] = (Y[train_to + (i * test_step)])
 
+colors = {'predictor' : colorset[0],
+          'naive' : colorset[1],
+          'true' : colorset[3]}
+
 time = range(0, len(y_hat))
-plt.plot(time, y, "g-", label="Measurement")
-plt.plot(time, y_hat, "r-", label="KNN Label")
-plt.plot(time, naive_hat, "b-", label="Naive Label")
+plt.plot(time, y, color=colors['true'], label="Measurement")
+plt.plot(time, y_hat, color=colors['predictor'], label="KNN-predicted")
+plt.plot(time, naive_hat, color=colors['naive'], label="Naive-predicted")
+plt.xlabel("Time [600s]")
+plt.ylabel("Power [MW]")
 plt.xlim([9600, 9750])
 plt.ylim([-30, 50])
 plt.legend()
 
 plot_scatter = plt.subplot(2, 2, 2)
-plt.title("Naive Label and True Measurement")
+plt.title("Naive-predicted and True Measurement")
 col = abs(y - naive_hat)
-plt.scatter(y, naive_hat, c=col, linewidth=0.0, cmap=plt.cm.jet)
-plt.xlabel("Y")
-plt.ylabel("Naive Label")
+plt.scatter(y, naive_hat, c=col, linewidth=0.0, cmap=cmap)
+plt.xlabel("True Measurement [MW]")
+plt.ylabel("Naive-predicted Measurement [MW]")
 plt.xlim([0, 30])
 plt.ylim([0, 30])
 
 plot_abs = plt.subplot(2, 2, 3)
 plt.title("Absolute Difference")
-plt.plot(time, (y_hat - y), "r-", label="KNN Label")
-plt.plot(time, (naive_hat - y), "b-", label="Naive Label")
+plt.plot(time, (y_hat - y), color=colors['predictor'], label="KNN-predicted")
+plt.plot(time, (naive_hat - y), color=colors['true'], label="Naive-predicted")
 plt.xlim([9600, 9750])
 plt.ylim([-20, 30])
+plt.xlabel("Time [600s]")
+plt.ylabel("Deviation of True Power [MW]")
 plt.legend()
 
 plot_scatter = plt.subplot(2, 2, 4)
-plt.title("KNN Label and True Measurement")
+plt.title("KNN-predicted and True Measurement")
 col = abs(y - y_hat)
-plt.scatter(y, y_hat, c=col, linewidth=0.0, cmap=plt.cm.jet)
-plt.xlabel("Y")
-plt.ylabel("KNN Label")
+plt.scatter(y, y_hat, c=col, linewidth=0.0, cmap=cmap)
+plt.xlabel("True Measurement [MW]")
+plt.ylabel("KNN-predicted Measurement [MW]")
 plt.xlim([0, 30])
 plt.ylim([0, 30])
 

examples/prediction/lin_regression_turbine.py

@@ -24,12 +24,10 @@
 
 import math
 import matplotlib.pyplot as plt
-
 from numpy import zeros, float32
 from windml.datasets.nrel import NREL
+from windml.visualization.colorset import cmap, colorset
 from windml.mapping.power_mapping import PowerMapping
-
-from sklearn.grid_search import GridSearchCV
 from sklearn import linear_model
 
 # get windpark and corresponding target. forecast is for the target turbine
@@ -82,47 +80,54 @@
 figure = plt.figure(figsize=(15, 10))
 
 plot_abs = plt.subplot(2, 2, 1)
-plt.title("Absolute Labels and True Measurements")
+plt.title("Predicted and True Measurements")
 
 # Array of true labels for plotting.
 y = zeros(len(y_hat))
 for i in range(0, len(y_hat)):
     y[i] = (Y[train_to + (i * test_step)])
 
+colors = {'predictor' : colorset[0],
+          'naive' : colorset[1],
+          'true' : colorset[3]}
+
 time = range(0, len(y_hat))
-plt.plot(time, y, "g-", label="Measurement")
-plt.plot(time, y_hat, "r-", label="Linear Label")
-plt.plot(time, naive_hat, "b-", label="Naive Label")
+plt.plot(time, y, color=colors['true'], label="Measurement")
+plt.plot(time, y_hat, color=colors['predictor'], label="Linear-predicted")
+plt.plot(time, naive_hat, color=colors['naive'], label="Naive-predicted")
+plt.xlabel("Time [600s]")
+plt.ylabel("Power [MW]")
 plt.xlim([9600, 9750])
 plt.ylim([-30, 50])
 plt.legend()
 
 plot_scatter = plt.subplot(2, 2, 2)
-plt.title("Naive Label and True Measurement")
+plt.title("Naive-predicted and True Measurement")
 col = abs(y - naive_hat)
-plt.scatter(y, naive_hat, c=col, linewidth=0.0, cmap=plt.cm.jet)
-plt.xlabel("Y")
-plt.ylabel("Naive Label")
+plt.scatter(y, naive_hat, c=col, linewidth=0.0, cmap=cmap)
+plt.xlabel("True Measurement [MW]")
+plt.ylabel("Naive-predicted Measurement [MW]")
 plt.xlim([0, 30])
 plt.ylim([0, 30])
 
 plot_abs = plt.subplot(2, 2, 3)
 plt.title("Absolute Difference")
-plt.plot(time, (y_hat - y), "r-", label="Linear Label")
-plt.plot(time, (naive_hat - y), "b-", label="Naive Label")
+plt.plot(time, (y_hat - y), color=colors['predictor'], label="Linear-predicted")
+plt.plot(time, (naive_hat - y), color=colors['true'], label="Naive-predicted")
 plt.xlim([9600, 9750])
 plt.ylim([-20, 30])
+plt.xlabel("Time [600s]")
+plt.ylabel("Deviation of True Power [MW]")
 plt.legend()
 
 plot_scatter = plt.subplot(2, 2, 4)
-plt.title("Linear Label and True Measurement")
+plt.title("Linear-predicted and True Measurement")
 col = abs(y - y_hat)
-plt.scatter(y, y_hat, c=col, linewidth=0.0, cmap=plt.cm.jet)
-plt.xlabel("Y")
-plt.ylabel("Linear Label")
+plt.scatter(y, y_hat, c=col, linewidth=0.0, cmap=cmap)
+plt.xlabel("True Measurement [MW]")
+plt.ylabel("Linear-predicted Measurement [MW]")
 plt.xlim([0, 30])
 plt.ylim([0, 30])
 
 plt.show()
 
-