# -*- coding: utf-8 -*-
"""
@author: Satoshi Hara
"""
import numpy as np
import pandas as pd
from collections import Counter
import glob
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.cm as cm
import matplotlib.colorbar as colorbar
import pylab as pl
class RTree(object):
def __init__(self, modeltype='regression'):
self.modeltype_ = modeltype
self.left_ = []
self.right_ = []
self.index_ = []
self.threshold_ = []
self.pred_ = []
def fit(self, filename):
df = pd.read_csv(filename, sep="\s+", skiprows=1, header=None)
for i in range(df.shape[0]):
if not int(df.ix[i, 5]) == -1:
left = int(df.ix[i, 1]) - 1
right = int(df.ix[i, 2]) - 1
index = int(df.ix[i, 3]) - 1
threshold = df.ix[i, 4]
pred = np.nan
else:
left = -1
right = -1
index = -1
threshold = 0
if self.modeltype_ == 'regression':
pred = df.ix[i, 6]
elif self.modeltype_ == 'classification':
pred = int(df.ix[i, 6]) - 1
self.left_.append(left)
self.right_.append(right)
self.index_.append(index)
self.threshold_.append(threshold)
self.pred_.append(pred)
def predict(self, X):
y = np.zeros(X.shape[0])
for n in range(X.shape[0]):
i = 0
while True:
if self.left_[i] == -1:
y[n] = self.pred_[i]
break
else:
if X[n, self.index_[i]] <= self.threshold_[i]:
i = self.left_[i]
else:
i = self.right_[i]
if self.modeltype_ == 'classification':
y = y.astype(np.int64)
return y
class RForest(object):
def __init__(self, modeltype='regression'):
self.modeltype_ = modeltype
self.trees_ = []
def fit(self, dirname):
self.trees_ = []
filenames = glob.glob(dirname + "*")
for filename in filenames:
tree = RTree(modeltype=self.modeltype_)
tree.fit(filename)
self.trees_.append(tree)
def predict(self, X):
Z = []
for tree in self.trees_:
Z.append(tree.predict(X))
Z = np.array(Z).T
if self.modeltype_ == 'regression':
y = np.mean(Z, axis=1)
elif self.modeltype_ == 'classification':
y = np.zeros(X.shape[0])
for n in range(X.shape[0]):
key = list(Counter(Z[n, :]).keys())
count = np.array(list(Counter(Z[n, :]).values()))
idx = np.argmax(count)
y[n] = key[idx]
y = y.astype(np.int64)
return y
def plot(self, X, d1, d2, alpha=0.8, treenum=5, filename=None, box0=None, plot_line=[]):
# boxes
num = X.shape[0]
dim = X.shape[1]
if box0 is None:
box0 = np.zeros((2, dim))
for d in range(dim):
box0[0, d] = np.min(X[:, d])
box0[1, d] = np.max(X[:, d])
box0 = box0.astype(np.float64)
boxes = []
for n in range(num):
box = box0.copy()
for j, tree in enumerate(self.trees_):
if j >= treenum:
break
i = 0
while True:
if tree.left_[i] == -1:
break
else:
if X[n, tree.index_[i]] <= tree.threshold_[i]:
box[1, tree.index_[i]] = min(box[1, tree.index_[i]], tree.threshold_[i])
i = tree.left_[i]
else:
box[0, tree.index_[i]] = max(box[0, tree.index_[i]], tree.threshold_[i])
i = tree.right_[i]
flg = True
for b in boxes:
flg *= np.max(np.abs(b[1] - box)) > 1e-8
if flg:
pred = self.predict(X[n, :][np.newaxis, :])[0]
boxes.append((pred, box))
n = len(boxes)
# plot
cmap = cm.get_cmap('cool')
fig, (ax1, ax2) = plt.subplots(1, 2, gridspec_kw = {'width_ratios':[19, 1]})
for b in boxes:
pred = b[0]
box = b[1]
c = cmap(1.0 * pred)
ax1.add_patch(pl.Rectangle(xy=[box[0, d1], box[0, d2]], width=(box[1, d1] - box[0, d1]), height=(box[1, d2] - box[0, d2]), facecolor=c, linewidth='1.0', alpha=alpha))
if len(plot_line) > 0:
ax1.plot(plot_line[0], plot_line[1], 'k--')
ax1.set_xlabel('x1', size=22)
ax1.set_ylabel('x2', size=22)
ax1.set_title('Learned Ensemble', size=28)
colorbar.ColorbarBase(ax2, cmap=cmap, format='%.1f')
ax2.set_ylabel('Predictor y', size=22)
plt.show()
if not filename is None:
plt.savefig(filename.replace('.pdf', '_%d.pdf' % (n,)), format="pdf", bbox_inches="tight")
plt.close()
