From 9bde30f6612448866e882edc86cb6b11da5b0c5e Mon Sep 17 00:00:00 2001
From: Nelly Barret <nelly.barret@etu.univ-lyon1.fr>
Date: Fri, 15 May 2020 19:37:44 +0200
Subject: [PATCH] [M] working on similarity between curves
---
predihood/charts.py | 6 +--
predihood/classes/Chart.py | 85 ++++++++++++++++++++++++++++++++++----
2 files changed, 80 insertions(+), 11 deletions(-)
diff --git a/predihood/charts.py b/predihood/charts.py
index d99c85aa..7fecdec2 100644
--- a/predihood/charts.py
+++ b/predihood/charts.py
@@ -8,12 +8,10 @@ def generate_charts():
data = Data(normalize="density", filter=True)
data.init_all_in_one()
lists = get_selected_indicators_lists(10)
- print(lists)
- for j, env in enumerate(["batiment", "usage"]):
- print(env)
+ for j, env in enumerate(["batiment"]):
dataset = Dataset(data, env, selected_indicators=lists["10"][env], train_size=0.8, test_size=0.2)
dataset.init_all_in_one()
- algo = Chart(name='chart', dataset=dataset)
+ algo = Chart(name='chart', dataset=dataset, number_of_iris=4)
algo.compute_trendline()
diff --git a/predihood/classes/Chart.py b/predihood/classes/Chart.py
index 8b566ab8..25159e59 100644
--- a/predihood/classes/Chart.py
+++ b/predihood/classes/Chart.py
@@ -6,37 +6,108 @@ from scipy.interpolate import interp1d
from predihood.classes.Method import Method
+def point_distance(y1, y2):
+ return abs(float(y2) - float(y1))
+
+
+def similarity_point(y1, y2, step):
+ distance = point_distance(y1, y2)
+
+ if distance == 0:
+ return 1 # points are the same
+ elif distance/(2*step) > 1:
+ return 0 # points are too different
+ else:
+ return distance/(2*step)
+
+
+def slope(x1, y1, x2, y2):
+ if y1 == 0 and y2 == 0: return 0
+ leading_coeff = (y2-y1)/(x2-x1) # coefficient directeur
+ # print(leading_coeff)
+ teta = np.degrees(np.arctan(leading_coeff)) # incline of the line between the two points
+ if x1 < x2 and y1 > y2: teta += 360
+ return teta
+
+
+def similarity(data1, data2, max_distance, nb_points):
+ somme = 0
+ for point1, point2 in zip(data1, data2):
+ x1, y1, x2, y2 = point1[0], point1[1], point2[0], point2[1]
+ similarity = point_distance(y1, y2) / max_distance
+ if x1+1 < len(data1) and x2+1 < len(data2):
+ next_x1 = data1[x1+1][0]
+ next_y1 = data1[x1+1][1]
+ next_x2 = data2[x2+1][0]
+ next_y2 = data2[x2+1][1]
+ print("next is (", next_x1, ";", next_y1, "), (", next_x2, ";", next_y2, ")")
+ sinus1 = np.sin(slope(x1, y1, next_x1, next_y1))
+ sinus2 = np.sin(slope(x2, y2, next_x2, next_y2))
+ slope_factor = abs(sinus1 - sinus2)
+ print(sinus1, "-", sinus2, "=", slope_factor)
+ else:
+ slope_factor = 0
+ similarity -= slope_factor
+ somme += similarity
+ somme /= nb_points
+ return somme
+
+
class Chart(Method):
- def __init__(self, name, dataset):
+ def __init__(self, name, dataset, number_of_iris=12):
Method.__init__(self, name, dataset)
self.chart = None
self.dataset = dataset
self.trendline = None
+ self.number_of_iris = number_of_iris if number_of_iris % 2 == 0 else 12
+ self.iris_per_line = 2
+ self.step = 0
def compute_trendline(self): # TODO: check order of selected indicators
print("compute trendline")
# for indicator in self.dataset.selected_indicators:
- fig, axs = plt.subplots(6, 2, figsize=(15, 15)) # rows, columns
+ fig, axs = plt.subplots(int(self.number_of_iris/2), self.iris_per_line, figsize=(15, 15)) # rows, columns
i, j, k = 0, 0, 1 # i and j are indices to plot sub-figures and k is the counter to place figures
- for index, row in self.dataset.data.head(12).iterrows():
+ for index, row in self.dataset.data.head(self.number_of_iris).iterrows():
data = []
for indicator in self.dataset.selected_indicators:
data.append(row[indicator])
- max_value = self.dataset.data.head(12)[self.dataset.selected_indicators].values.max()
+ max_value = self.dataset.data.head(self.number_of_iris)[self.dataset.selected_indicators].values.max()
x = np.arange(0, len(data))
y = data
f = interp1d(x, y)
axs[i, j].axis(ymin=0, ymax=max_value)
axs[i, j].set_xticks(np.arange(0, len(data)))
- axs[i, j].set_xticks(np.arange(0, max_value, step=max_value / 5))
+ self.step = max_value/5
+ axs[i, j].set_xticks(np.arange(0, max_value, step=self.step))
axs[i, j].plot(x, data, 'o', x, f(x), '-')
title = str(row['CODE']) + " - " + str(self.dataset.env)
axs[i, j].set_title(title)
- if k < 2:
+ if k < self.iris_per_line:
k += 1
j += 1
else:
k = 1
i += 1
j = 0
- fig.show()
\ No newline at end of file
+ fig.show()
+ self.compute_similarity()
+
+ def compute_similarity(self):
+ for index1, row1 in self.dataset.data.head(self.number_of_iris).iterrows():
+ data1 = []
+ data2 = []
+ for index2, row2 in self.dataset.data.head(self.number_of_iris).iterrows():
+ if index1 > index2: # compare charts only once (don't compare 2 and 1 and after 1 and 2)
+ for i in range(len(self.dataset.selected_indicators)):
+ indicator = self.dataset.selected_indicators[i]
+ data1.append((i, row1[indicator]))
+ data2.append((i, row2[indicator]))
+ sim_percentage = similarity(data1, data2, self.step, len(self.dataset.selected_indicators))
+ print(row1['CODE'], "/", row2['CODE'], " ->", sim_percentage)
+ print(data1)
+ print(data2)
+
+
+if __name__ == '__main__':
+ print(slope(0, 1, 1, 0))
--
GitLab