32 KiB
32 KiB
In [1]:
## USE for Multi GPU Systems #import os #os.environ["CUDA_VISIBLE_DEVICES"]="0" %matplotlib inline from scipy.odr import * from scipy.stats import * import numpy as np import pandas as pd import os import time import matplotlib.pyplot as plt import ast from multiprocessing import Pool import scipy from IPython import display from matplotlib.patches import Rectangle from sklearn.metrics import mean_squared_error import json import scipy.stats as st from sklearn.metrics import r2_score from matplotlib import cm from mpl_toolkits.mplot3d import axes3d import matplotlib.pyplot as plt from matplotlib.patches import Ellipse import copy from sklearn.model_selection import LeaveOneOut, LeavePOut from multiprocessing import Pool import cv2 import sklearn import random from sklearn import neighbors from sklearn import svm from sklearn import tree from sklearn import ensemble from sklearn.model_selection import GridSearchCV from sklearn.metrics import classification_report import numpy as np import matplotlib.pyplot as plt import pandas as pd import math # Importing matplotlib to plot images. import matplotlib.pyplot as plt import numpy as np %matplotlib inline # Importing SK-learn to calculate precision and recall import sklearn from sklearn import metrics from sklearn.model_selection import train_test_split, cross_val_score, LeaveOneGroupOut from sklearn.utils import shuffle from sklearn.model_selection import GridSearchCV from sklearn.metrics.pairwise import euclidean_distances from sklearn.metrics import confusion_matrix from sklearn.metrics import accuracy_score import pickle as pkl import h5py from pathlib import Path import os.path import sys import datetime import time import skimage target_names = ["Knuckle", "Finger"]
In [2]:
from skimage import measure from skimage.measure import find_contours, approximate_polygon, \ subdivide_polygon, EllipseModel, LineModelND
In [3]:
def getEllipseParams(img): points = np.argwhere(img > 40) contours = skimage.measure.find_contours(img, 40) points_to_approx = [] highest_val = 0 for n, contour in enumerate(contours): if (len(contour) > highest_val): points_to_approx = contour highest_val = len(contour) try: contour = np.fliplr(points_to_approx) except Exception as inst: return [-1, -1, -1, -1, -1] ellipse = skimage.measure.fit.EllipseModel() ellipse.estimate(contour) try: xc, yc, a, b, theta = ellipse.params except Exception as int: return [-1, -1, -1, -1, -1] return [xc, yc, a, b, theta]
In [4]:
# the data, split between train and test sets df = pd.read_pickle("DataStudyCollection/df_statistics.pkl") lst = df.userID.unique() np.random.seed(42) np.random.shuffle(lst) test_ids = lst[-5:] train_ids = lst[:-5] df["Set"] = "Test" df.loc[df.userID.isin(train_ids), "Set"] = "Train" print(train_ids, test_ids) print(len(train_ids), ":", len(test_ids)) print(len(train_ids) / len(lst), ":", len(test_ids)/ len(lst)) #df_train = df[df.userID.isin(train_ids)] #df_test = df[df.userID.isin(test_ids) & (df.Version == "Normal")]
[ 1 2 9 6 4 14 17 16 12 3 10 18 5] [13 8 11 15 7] 13 : 5 0.7222222222222222 : 0.2777777777777778
In [5]:
fig, ax = plt.subplots(1) img = df.iloc[0].Blobs xc, yc, a, b, theta = getEllipseParams(img) ax.imshow(img) e = Ellipse(xy=[xc,yc], width=a*2, height=b*2, angle=math.degrees(theta), fill=False, lw=2, edgecolor='w') ax.add_artist(e)
Out[5]:
<matplotlib.patches.Ellipse at 0x7ff60430b668>
In [6]:
lst = df.Blobs.apply(lambda x: getEllipseParams(x))
In [7]:
lst2 = np.vstack(lst.values)
In [8]:
lst2.shape
Out[8]:
(618012, 5)
In [9]:
df["XC"] = lst2[:,0] df["YC"] = lst2[:,1] df["EllipseW"] = lst2[:,2] df["EllipseH"] = lst2[:,3] df["EllipseTheta"] = lst2[:,4]
In [10]:
df["Area"] = df["EllipseW"] * df["EllipseH"] * np.pi df["AvgCapa"] = df.Blobs.apply(lambda x: np.mean(x)) df["SumCapa"] = df.Blobs.apply(lambda x: np.sum(x))
In [11]:
lst = list(range(1, df.userID.max())) SEED = 42#448 random.seed(SEED) random.shuffle(lst) lst
Out[11]:
[8, 11, 6, 7, 16, 15, 14, 10, 9, 2, 3, 13, 17, 5, 12, 1, 4]
In [12]:
dfY = df[df.Set == "Train"].copy(deep=True) dfT = df[(df.Set == "Test") & (df.Version == "Normal")].copy(deep=True)
In [13]:
minmax = min(len(dfY[dfY.Input == "Finger"]), len(dfY[dfY.Input == "Knuckle"])) dfX = dfY[dfY.Input == "Finger"].sample(minmax) dfZ = dfY[dfY.Input == "Knuckle"].sample(minmax) dfY = pd.concat([dfX,dfZ]) minmax = min(len(dfT[dfT.Input == "Finger"]), len(dfT[dfT.Input == "Knuckle"])) dfX = dfT[dfT.Input == "Finger"].sample(minmax) dfZ = dfT[dfT.Input == "Knuckle"].sample(minmax) dfT = pd.concat([dfX,dfZ])
In [14]:
dfT.groupby("Input").count()
Out[14]:
| userID | Timestamp | Current_Task | Task_amount | TaskID | VersionID | RepetitionID | Actual_Data | Is_Pause | Image | ... | InputMethod | Set | XC | YC | EllipseW | EllipseH | EllipseTheta | Area | AvgCapa | SumCapa | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Input | |||||||||||||||||||||
| Finger | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | ... | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 |
| Knuckle | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | ... | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 | 9421 |
2 rows × 31 columns
FEATURE SET: sum of capacitance, avg of capacitance, ellipse area, ellipse width, height and theta.¶
In [15]:
features = ["SumCapa", "AvgCapa", "Area", "EllipseW", "EllipseH", "EllipseTheta"]
ZeroR¶
In [16]:
dfT["InputMethodPred"] = 1
In [17]:
print(confusion_matrix(dfT.InputMethod.values, dfT.InputMethodPred.values, labels=[0, 1])) print("Accuray: %.2f" % accuracy_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Recall: %.2f" % metrics.recall_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print("Precision: %.2f" % metrics.average_precision_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print("F1-Score: %.2f" % metrics.f1_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print(classification_report(dfT.InputMethod.values, dfT.InputMethodPred.values, target_names=target_names))
[[ 0 9421]
[ 0 9421]]
Accuray: 0.50
Recall: 0.50
Precision: 0.50
F1-Score: 0.33
precision recall f1-score support
Knuckle 0.00 0.00 0.00 9421
Finger 0.50 1.00 0.67 9421
micro avg 0.50 0.50 0.50 18842
macro avg 0.25 0.50 0.33 18842
weighted avg 0.25 0.50 0.33 18842
/usr/local/lib/python3.6/dist-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 in labels with no predicted samples. 'precision', 'predicted', average, warn_for) /usr/local/lib/python3.6/dist-packages/sklearn/metrics/classification.py:1143: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. 'precision', 'predicted', average, warn_for)
DecisionTreeClassifier¶
In [18]:
%%time param_grid = {'max_depth': range(2,32,1), 'min_samples_split':range(2,10,1)} #TODO: Create Baseline for different ML stuff clf = GridSearchCV(tree.DecisionTreeClassifier(), param_grid, cv=5 , n_jobs=os.cpu_count()-2, verbose=1) clf.fit(dfY[features].values, dfY.InputMethod.values) print(clf.best_params_, clf.best_score_) dfT["InputMethodPred"] = clf.predict(dfT[features].values) print(confusion_matrix(dfT.InputMethod.values, dfT.InputMethodPred.values, labels=[0, 1])) print("Accuray: %.3f" % accuracy_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Recall: %.3f" % metrics.recall_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print("Precision: %.3f" % metrics.average_precision_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print("F1-Score: %.3f" % metrics.f1_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print(classification_report(dfT.InputMethod.values, dfT.InputMethodPred.values, target_names=target_names))
Fitting 5 folds for each of 240 candidates, totalling 1200 fits
[Parallel(n_jobs=30)]: Using backend LokyBackend with 30 concurrent workers. [Parallel(n_jobs=30)]: Done 140 tasks | elapsed: 10.4s [Parallel(n_jobs=30)]: Done 390 tasks | elapsed: 31.4s [Parallel(n_jobs=30)]: Done 740 tasks | elapsed: 1.3min [Parallel(n_jobs=30)]: Done 1200 out of 1200 | elapsed: 2.4min finished
{'max_depth': 22, 'min_samples_split': 2} 0.8120637794585754
[[7409 2012]
[3096 6325]]
Accuray: 0.73
Recall: 0.73
Precision: 0.67
F1-Score: 0.73
precision recall f1-score support
Knuckle 0.71 0.79 0.74 9421
Finger 0.76 0.67 0.71 9421
micro avg 0.73 0.73 0.73 18842
macro avg 0.73 0.73 0.73 18842
weighted avg 0.73 0.73 0.73 18842
CPU times: user 7.26 s, sys: 3.38 s, total: 10.6 s
Wall time: 2min 29s
RandomForestClassifier¶
In [19]:
%%time param_grid = {'n_estimators': range(55,64,1), 'max_depth': range(50,70,1)} #TODO: Create Baseline for different ML stuff clf = GridSearchCV(ensemble.RandomForestClassifier(), param_grid, cv=5 , n_jobs=os.cpu_count()-2, verbose=1) clf.fit(dfY[features].values, dfY.InputMethod.values) print(clf.best_params_, clf.best_score_) dfT["InputMethodPred"] = clf.predict(dfT[features].values) print(confusion_matrix(dfT.InputMethod.values, dfT.InputMethodPred.values, labels=[0, 1])) print("Accuray: %.2f" % accuracy_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Recall: %.2f" % metrics.recall_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Precision: %.2f" % metrics.average_precision_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("F1-Score: %.2f" % metrics.f1_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print(classification_report(dfT.InputMethod.values, dfT.InputMethodPred.values, target_names=target_names))
Fitting 5 folds for each of 180 candidates, totalling 900 fits
[Parallel(n_jobs=94)]: Using backend LokyBackend with 94 concurrent workers. [Parallel(n_jobs=94)]: Done 12 tasks | elapsed: 1.2min [Parallel(n_jobs=94)]: Done 262 tasks | elapsed: 4.0min [Parallel(n_jobs=94)]: Done 612 tasks | elapsed: 9.2min [Parallel(n_jobs=94)]: Done 900 out of 900 | elapsed: 12.8min finished
{'max_depth': 60, 'n_estimators': 63} 0.8669582104371696
[[8175 1246]
[2765 6656]]
Accuray: 0.79
Recall: 0.71
Precision: 0.74
F1-Score: 0.77
precision recall f1-score support
Knuckle 0.75 0.87 0.80 9421
Finger 0.84 0.71 0.77 9421
micro avg 0.79 0.79 0.79 18842
macro avg 0.79 0.79 0.79 18842
weighted avg 0.79 0.79 0.79 18842
CPU times: user 42.1 s, sys: 834 ms, total: 42.9 s
Wall time: 13min 28s
kNN¶
In [20]:
%%time param_grid = {'n_neighbors': range(2,64,1), #weights': ['uniform', 'distance'] } #TODO: Create Baseline for different ML stuff clf = GridSearchCV(neighbors.KNeighborsClassifier(), param_grid, cv=5 , n_jobs=os.cpu_count()-2, verbose=1) clf.fit(dfY[features].values, dfY.InputMethod.values) print(clf.best_params_, clf.best_score_) dfT["InputMethodPred"] = clf.predict(dfT[features].values) print(confusion_matrix(dfT.InputMethod.values, dfT.InputMethodPred.values, labels=[0, 1])) print("Accuray: %.2f" % accuracy_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Recall: %.2f" % metrics.recall_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print("Precision: %.2f" % metrics.average_precision_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print("F1-Score: %.2f" % metrics.f1_score(dfT.InputMethod.values, dfT.InputMethodPred.values, average="macro")) print(classification_report(dfT.InputMethod.values, dfT.InputMethodPred.values, target_names=target_names))
Fitting 5 folds for each of 62 candidates, totalling 310 fits
[Parallel(n_jobs=94)]: Using backend LokyBackend with 94 concurrent workers. [Parallel(n_jobs=94)]: Done 12 tasks | elapsed: 17.7s [Parallel(n_jobs=94)]: Done 310 out of 310 | elapsed: 1.5min finished
{'n_neighbors': 2} 0.800546827088748
[[8187 1234]
[4318 5103]]
Accuray: 0.71
Recall: 0.54
Precision: 0.67
F1-Score: 0.65
precision recall f1-score support
Knuckle 0.65 0.87 0.75 9421
Finger 0.81 0.54 0.65 9421
micro avg 0.71 0.71 0.71 18842
macro avg 0.73 0.71 0.70 18842
weighted avg 0.73 0.71 0.70 18842
CPU times: user 1.74 s, sys: 300 ms, total: 2.04 s
Wall time: 1min 30s
SVM¶
In [21]:
%%time C_range = np.logspace(1, 3,3) gamma_range = np.logspace(-1, 1, 3) param_grid = dict(gamma=gamma_range, C=C_range) clf = GridSearchCV(sklearn.svm.SVC(), param_grid, cv=5 , n_jobs=os.cpu_count()-2, verbose=1) clf.fit(dfY[features].values, dfY.InputMethod.values) print(clf.best_params_, clf.best_score_) dfT["InputMethodPred"] = clf.predict(dfT[features].values)
Fitting 5 folds for each of 9 candidates, totalling 45 fits
[Parallel(n_jobs=94)]: Using backend LokyBackend with 94 concurrent workers. [Parallel(n_jobs=94)]: Done 42 out of 45 | elapsed: 1056.5min remaining: 75.5min [Parallel(n_jobs=94)]: Done 45 out of 45 | elapsed: 1080.5min finished
{'C': 10.0, 'gamma': 10.0} 0.8256943024851795
CPU times: user 2h 42min 9s, sys: 23.6 s, total: 2h 42min 33s
Wall time: 20h 43min 1s
In [22]:
print(clf.best_params_, clf.best_score_) print(confusion_matrix(dfT.InputMethod.values, dfT.InputMethodPred.values, labels=[0, 1])) print("Accuray: %.2f" % accuracy_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Recall: %.2f" % metrics.recall_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("Precision: %.2f" % metrics.average_precision_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print("F1-Score: %.2f" % metrics.f1_score(dfT.InputMethod.values, dfT.InputMethodPred.values)) print(classification_report(dfT.InputMethod.values, dfT.InputMethodPred.values, target_names=target_names))
{'C': 10.0, 'gamma': 10.0} 0.8256943024851795
[[7106 2315]
[2944 6477]]
Accuray: 0.72
Recall: 0.69
Precision: 0.66
F1-Score: 0.71
precision recall f1-score support
Knuckle 0.71 0.75 0.73 9421
Finger 0.74 0.69 0.71 9421
micro avg 0.72 0.72 0.72 18842
macro avg 0.72 0.72 0.72 18842
weighted avg 0.72 0.72 0.72 18842