Merge pull request #11534 from khs26/numpify-andrews-curves

PERF: Updated andrews_curves to use Numpy arrays for its samples

Author: jreback

asv_bench/benchmarks/plotting.py

@@ -2,9 +2,9 @@
 try:
     from pandas import date_range
 except ImportError:
-
     def date_range(start=None, end=None, periods=None, freq=None):
         return DatetimeIndex(start, end, periods=periods, offset=freq)
+from pandas.tools.plotting import andrews_curves
 
 
 class plot_timeseries_period(object):
@@ -16,4 +16,17 @@ def setup(self):
         self.df = DataFrame(np.random.randn(self.N, self.M), index=date_range('1/1/1975', periods=self.N))
 
     def time_plot_timeseries_period(self):
-        self.df.plot()
+        self.df.plot()
+
+class plot_andrews_curves(object):
+    goal_time = 0.6
+
+    def setup(self):
+        self.N = 500
+        self.M = 10
+	data_dict = {x: np.random.randn(self.N) for x in range(self.M)}
+	data_dict["Name"] = ["A"] * self.N
+        self.df = DataFrame(data_dict)
+
+    def time_plot_andrews_curves(self):
+        andrews_curves(self.df, "Name")

doc/source/whatsnew/v0.18.0.txt

@@ -91,7 +91,7 @@ Removal of prior version deprecations/changes
 Performance Improvements
 ~~~~~~~~~~~~~~~~~~~~~~~~
 
-
+- Improved performance of ``andrews_curves`` (:issue:`11534`)
 
 
 

pandas/tests/test_graphics_others.py

@@ -463,6 +463,26 @@ def test_andrews_curves(self):
         cmaps = lmap(cm.jet, np.linspace(0, 1, df['Name'].nunique()))
         self._check_colors(ax.get_lines()[:10], linecolors=cmaps, mapping=df['Name'][:10])
 
+        length = 10
+        df = DataFrame({"A": random.rand(length),
+                        "B": random.rand(length),
+                        "C": random.rand(length),
+                        "Name": ["A"] * length})
+
+        _check_plot_works(andrews_curves, frame=df, class_column='Name')
+
+        rgba = ('#556270', '#4ECDC4', '#C7F464')
+        ax = _check_plot_works(andrews_curves, frame=df, class_column='Name', color=rgba)
+        self._check_colors(ax.get_lines()[:10], linecolors=rgba, mapping=df['Name'][:10])
+
+        cnames = ['dodgerblue', 'aquamarine', 'seagreen']
+        ax = _check_plot_works(andrews_curves, frame=df, class_column='Name', color=cnames)
+        self._check_colors(ax.get_lines()[:10], linecolors=cnames, mapping=df['Name'][:10])
+
+        ax = _check_plot_works(andrews_curves, frame=df, class_column='Name', colormap=cm.jet)
+        cmaps = lmap(cm.jet, np.linspace(0, 1, df['Name'].nunique()))
+        self._check_colors(ax.get_lines()[:10], linecolors=cmaps, mapping=df['Name'][:10])
+
         colors = ['b', 'g', 'r']
         df = DataFrame({"A": [1, 2, 3],
                         "B": [1, 2, 3],

pandas/tools/plotting.py

@@ -507,6 +507,15 @@ def normalize(series):
 def andrews_curves(frame, class_column, ax=None, samples=200, color=None,
                    colormap=None, **kwds):
     """
+    Generates a matplotlib plot of Andrews curves, for visualising clusters of multivariate data.
+
+    Andrews curves have the functional form:
+
+    f(t) = x_1/sqrt(2) + x_2 sin(t) + x_3 cos(t) + x_4 sin(2t) + x_5 cos(2t) + ...
+
+    Where x coefficients correspond to the values of each dimension and t is linearly spaced between -pi and +pi. Each
+    row of frame then corresponds to a single curve.
+
     Parameters:
     -----------
     frame : DataFrame
@@ -527,28 +536,34 @@ def andrews_curves(frame, class_column, ax=None, samples=200, color=None,
     ax: Matplotlib axis object
 
     """
-    from math import sqrt, pi, sin, cos
+    from math import sqrt, pi
     import matplotlib.pyplot as plt
 
     def function(amplitudes):
-        def f(x):
+        def f(t):
             x1 = amplitudes[0]
             result = x1 / sqrt(2.0)
-            harmonic = 1.0
-            for x_even, x_odd in zip(amplitudes[1::2], amplitudes[2::2]):
-                result += (x_even * sin(harmonic * x) +
-                           x_odd * cos(harmonic * x))
-                harmonic += 1.0
-            if len(amplitudes) % 2 != 0:
-                result += amplitudes[-1] * sin(harmonic * x)
+
+            # Take the rest of the coefficients and resize them appropriately. Take a copy of amplitudes as otherwise
+            # numpy deletes the element from amplitudes itself.
+            coeffs = np.delete(np.copy(amplitudes), 0)
+            coeffs.resize((coeffs.size + 1) / 2, 2)
+
+            # Generate the harmonics and arguments for the sin and cos functions.
+            harmonics = np.arange(0, coeffs.shape[0]) + 1
+            trig_args = np.outer(harmonics, t)
+
+            result += np.sum(coeffs[:, 0, np.newaxis] * np.sin(trig_args) +
+                             coeffs[:, 1, np.newaxis] * np.cos(trig_args),
+                             axis=0)
             return result
         return f
 
     n = len(frame)
     class_col = frame[class_column]
     classes = frame[class_column].drop_duplicates()
     df = frame.drop(class_column, axis=1)
-    x = [-pi + 2.0 * pi * (t / float(samples)) for t in range(samples)]
+    t = np.linspace(-pi, pi, samples)
     used_legends = set([])
 
     color_values = _get_standard_colors(num_colors=len(classes),
@@ -560,14 +575,14 @@ def f(x):
     for i in range(n):
         row = df.iloc[i].values
         f = function(row)
-        y = [f(t) for t in x]
+        y = f(t)
         kls = class_col.iat[i]
         label = com.pprint_thing(kls)
         if label not in used_legends:
             used_legends.add(label)
-            ax.plot(x, y, color=colors[kls], label=label, **kwds)
+            ax.plot(t, y, color=colors[kls], label=label, **kwds)
         else:
-            ax.plot(x, y, color=colors[kls], **kwds)
+            ax.plot(t, y, color=colors[kls], **kwds)
 
     ax.legend(loc='upper right')
     ax.grid()