Adding visualizing pca reconstruction. packages required: numpy, skle…

…arn, matplotlib

visualization/pca_reconstruction.py

@@ -0,0 +1,75 @@
+import matplotlib.pyplot as plt
+from matplotlib import cm
+import numpy as np
+from sklearn import decomposition
+import csv
+
+
+def read_data(filname, limit=None):
+    data = []
+    labels = []
+
+    csv_reader = csv.reader(open(filname, "r"), delimiter=",")
+    index = 0
+    for row in csv_reader:
+        index += 1
+        if index == 1:
+            continue
+
+        labels.append(int(row[0]))
+        row = row[1:]
+
+        data.append(np.array(np.int64(row)))
+
+        if limit != None and index == limit + 1:
+            break
+
+    return (data, labels)
+
+print "Reading train data"
+train, target = read_data("../data/train.csv")
+
+pca_components = [1, 2, 3, 4, 5, 10, 20, 25, 30, 50, 70, 100]
+pca_fits = []
+
+for comp_size in pca_components:
+    print "Fitting pca with %d components" % comp_size
+    pca_fits.append(decomposition.PCA(n_components=comp_size).fit(train))
+
+figure = plt.figure()
+
+t = np.array(target)
+
+choosen_numbers = []
+
+choosen_numbers.append(np.argwhere(t == 0)[-3])
+choosen_numbers.append(np.argwhere(t == 1)[-3])
+choosen_numbers.append(np.argwhere(t == 2)[-3])
+choosen_numbers.append(np.argwhere(t == 3)[-3])
+choosen_numbers.append(np.argwhere(t == 4)[-3])
+choosen_numbers.append(np.argwhere(t == 5)[-3])
+choosen_numbers.append(np.argwhere(t == 6)[-3])
+choosen_numbers.append(np.argwhere(t == 7)[-3])
+choosen_numbers.append(np.argwhere(t == 8)[-3])
+choosen_numbers.append(np.argwhere(t == 9)[-3])
+
+pca_index = 1
+for n in choosen_numbers:
+    for p in pca_fits:
+        transformed = p.transform(train[n])
+        # print "Shape of transformed: %d" % transformed.shape
+        reconstructed = p.inverse_transform(transformed)
+        f = figure.add_subplot(10, len(pca_components), pca_index).imshow(np.reshape(reconstructed, (28, 28)), interpolation='nearest', cmap=cm.hot)  # cmap=cm.Greys_r)
+        for xlabel_i in f.axes.get_xticklabels():
+            xlabel_i.set_visible(False)
+            xlabel_i.set_fontsize(0.0)
+        for xlabel_i in f.axes.get_yticklabels():
+            xlabel_i.set_fontsize(0.0)
+            xlabel_i.set_visible(False)
+        for tick in f.axes.get_xticklines():
+            tick.set_visible(False)
+        for tick in f.axes.get_yticklines():
+            tick.set_visible(False)
+        pca_index += 1
+
+plt.show()