notebook.community

Edit and run



In [1]:

    
import pandas as pd
import numpy as np



In [2]:

    
train = pd.read_csv('RS.txt', names=['ID','Timestep','Front','Rear'])
train.head()



In [3]:

    
labels = np.zeros(len(train))
startw = [461,  1795, 3424, 5023, 6733, 8391]
endw =   [1502, 2275, 4450, 6074, 7760, 9436]
for s,e in zip(startw,endw):
    labels[range(s,e)] = 1



In [4]:

    
train['Label'] = pd.Series(labels, index=train.index)



In [5]:

    
import matplotlib.pyplot as plt
%matplotlib inline



In [21]:

    
plt.figure(figsize=(10,8))
plt.subplot(221)
plt.plot(train.Front)
plt.subplot(222)
plt.plot(train.Rear)
plt.subplot(223)
plt.plot(train.Label)









    Out[21]:





[<matplotlib.lines.Line2D at 0x10b843f10>]

Existing labels

Note: 2275 - ~3000 doesn't look like 'stopped'!



In [23]:

    
plt.scatter(train.Timestep, 
            train.Front, 
            marker='.', 
            linewidth=0, 
            c=['blue' if x == 0 else 'green' for x in train.Label])









    Out[23]:





<matplotlib.collections.PathCollection at 0x10c07f6d0>



In [8]:

    
from sklearn.preprocessing import StandardScaler



In [9]:

    
training_features = [np.append(abs(np.fft.fft(train.Front[x:x+60]))[0:6], abs(np.fft.fft(train.Rear[x:x+60]))[0:6]) for x in range(0, len(train)-60)]



In [10]:

    
scaler = StandardScaler()



In [11]:

    
from sklearn.neighbors import KNeighborsClassifier
knn = KNeighborsClassifier(n_neighbors=11, weights='uniform')
knn.fit(scaler.fit_transform(training_features), train.Label[60:])









    Out[11]:





KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
           metric_params=None, n_neighbors=11, p=2, weights='uniform')



In [12]:

    
test = pd.read_csv('RS_test.txt', names=['ID','Timestep','Front','Rear'])
test.head()



In [13]:

    
testing_features = [np.append(abs(np.fft.fft(test.Front[x:x+60]))[0:6], abs(np.fft.fft(test.Rear[x:x+60]))[0:6]) for x in range(0, len(test)-60)]



In [14]:

    
test_prediction = knn.predict(scaler.fit_transform(testing_features))



In [15]:

    
plt.subplot(221)
plt.plot(test.Front)
plt.subplot(222)
plt.plot(test.Rear)









    Out[15]:





[<matplotlib.lines.Line2D at 0x10a957f90>]



In [16]:

    
test_points = test[60:]

Result of prediction



In [17]:

    
plt.figure(figsize=(12,8))
plt.scatter(test_points.Timestep, test_points.Front, marker='.', linewidth=0, c=['blue' if x == 0 else 'green' for x in test_prediction])









    Out[17]:





<matplotlib.collections.PathCollection at 0x10ac09c10>

	ID	Timestep	Front	Rear
0	1	1820	408	80
1	1	1821	409	80
2	1	1822	409	81
3	1	1823	408	80
4	1	1824	407	83