def RF_model(X, y):
RF_clf = RandomForestClassifier(n_estimators=150, max_depth=10, min_samples_split=100,
min_samples_leaf=100, n_jobs=-1)
# 再度分割数据集, 取1/3数据用作训练集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.66, random_state=42)
# 拟合模型
RF_clf.fit(X_train, y_train)
# 做预测
y_pred = RF_clf.predict(X_test)
print("Accuracy: {0:.3f}\n".format(accuracy_score(y_test, y_pred)))
print('Kappa:', skll.kappa(y_test, y_pred))
print("\nConfusion matrix:\n{}\n".format(confusion_matrix(y_test, y_pred)))
labels = np.unique(np.concatenate((y_test, y_pred)))
scores_image = mglearn.tools.heatmap(
confusion_matrix(y_test, y_pred), xlabel='Predicted label',
ylabel='True label', xticklabels=labels, yticklabels=labels,
cmap='viridis', fmt='%d')
plt.title("Confusion matrix")
plt.gca().invert_yaxis()
print(classification_report(y_test, y_pred))
print("The training of the model ends!")
return RF_clf
def kappa_score(events, data_student, data_RE):
for event in events:
label_student = check_label_event(event, data_student, 'student')
label_RE = check_label_event(event, data_RE, 'RE')
print event + '\t' + str(kappa(label_RE, label_student))
def cal_cohen_kappa_by_hit(results):
# collect predicate by hit_id
preds_by_hit = {}
for sent in results.itervalues():
for pred in sent.predicates.itervalues():
if pred.hit_id not in preds_by_hit:
preds_by_hit[pred.hit_id] = []
preds_by_hit[pred.hit_id].append(pred)
kappas = []
# calculate kappa
for preds in preds_by_hit.itervalues():
# get answer list for each worker
worker_answers_by_hit = {worker_id: [] for worker_id
in preds[0].workers.keys()}
for pred in preds:
for worker in pred.workers.itervalues():
worker_answers_by_hit[worker.workerid].append(worker.answer)
# calculate kappa for each answer list pair
worker_answers = worker_answers_by_hit.values()
pairs_combinations = itertools.combinations(worker_answers, 2)
for l1, l2 in pairs_combinations:
kappa_value = skll.kappa(l1, l2) #, weights='quadratic')
kappas.append(kappa_value)
avg_kappa = sum(kappas) / (len(kappas) + 0.0)
print len(preds_by_hit)
print "The average Cohen's kappa of all HITs is %f." %(avg_kappa)
def kapp_sentiment_event(path_sent, path_event, sentiment, event):
load_sent, load_event = load_file(path_sent, 'allTweets_ver3_sentLabel_' + sentiment + '.csv')\
, load_file(path_event, event + '.csv')
list_sent, list_gt = list(), list()
for i in range(0, len(load_sent)):
split_sent, split_gt = load_sent[i].split('\t'), load_event[i].split('\t')
label, gt = int(split_sent[0]), int(split_gt[1])
list_sent.append(label), list_gt.append(gt)
print 'Kappa score of ' + sentiment + ' and ' + event + ':' + '\t' + str(kappa(list_gt, list_sent))
def Kappa(y_true, y_pred, **kwargs):
if not KAGGLE:
from skll import kappa
#from kappa import kappa
return kappa(y_true, y_pred, **kwargs)
ftTest = d.transform(test.T.to_dict().values())
msk = np.random.rand(ftData.shape[0]) < 0.8
ftTrain = ftData[msk, 0:ftData.shape[1]]
ftCv = ftData[~msk, 0:ftData.shape[1]]
yTrain = dataPred[msk]
yCv = dataPred[~msk]
clf = GradientBoostingRegressor(n_estimators=3000,
max_depth=8,
min_samples_split=10,
min_samples_leaf=2,
max_features="auto",
verbose=1,
random_state=1988)
clf.fit(ftTrain, yTrain)
trainPred = clf.predict(ftTrain)
trainPred = [adjustResponse(resp) for resp in trainPred]
kTrain = kappa(yTrain, trainPred, weights="quadratic")
print "Ktrain : " + str(kTrain)
cvPred = clf.predict(ftCv)
cvPred = [adjustResponse(resp) for resp in cvPred]
kCv = kappa(yCv, cvPred, weights="quadratic")
print "cvPred : " + str(kCv)
joblib.dump(clf, "GBMModel3000_sameCvMat/GBMModel3000_sameCvMat")
if datum['kevin']!='' and datum['amy']!='']) filtered_kevin,filtered_amy = zip(*[(int(datum['kevin']),int(datum['amy'])) for datum in filtered_data if datum['kevin']!='' and datum['amy']!='']) ratings = map(str,range(4)) kevin_amy_contingency = np.array([[len([item for item in filtered_data if item['kevin'] == kevin_rating and item['amy'] == amy_rating]) for kevin_rating in ratings] for amy_rating in ratings]) print>>f, 'Did Kevin and Amy rate differently? %s'%fisher_test(numpy2ri(kevin_amy_contingency),simulate_p_value=True,B=6000) print>>f, '---For all data------' print>>f, 'Chance agreement: %.02f'%(np.trace(contingency_table)/float(contingency_table.sum())) print>>f, "Cohen's kappa: %.02f"%skll.kappa(kevin,amy) print>>f, '---------' print>>f, '---For full fields-----' print>>f, 'Chance agreement: %.02f'%(np.trace(filtered_contingency_table)/float(filtered_contingency_table.sum())) print>>f, "Cohen's kappa: %.02f"%skll.kappa(kevin,amy) print>>f, '---------' #--Does reflection related to grade? unfiltered_distribution_grades = [float(item['exam']) for item in data if item['exam'] != ''] filtered_distribution_grades = [float(item['exam']) for item in filtered_data if item['exam'] != ''] quartiles = [(0,70),(70,80),(80,90),(90,100)] unfiltered_quartiles = [(np.percentile(unfiltered_distribution_grades,lower),np.percentile(unfiltered_distribution_grades,upper))
data = []
sentiment_r1_5_scale = []
sentiment_r2_5_scale = []
for r1, r2 in zip(sentences_r1, sentences_r2):
sentiment_r1_5_scale.append(int(r1[5]))
data.append((6, r1[0], r1[5]))
sentiment_r2_5_scale.append(int(r2[5]))
data.append((7, r2[0], r2[5]))
if (r1[0] != r2[0]):
print r1[0]
except Exception, e:
print e
# disconnect from server
db.close()
print i
print skll.kappa(sentiment_r1_5_scale, sentiment_r2_5_scale)
annotation = AnnotationTask(data=data)
print annotation.kappa()
print annotation.alpha()
def Kappa(y_true, y_pred, **kwargs):
if not KAGGLE:
from skll import kappa
#from kappa import kappa
return kappa(y_true, y_pred, **kwargs)
def evalerror(preds, dtrain):
labels = dtrain.get_label()
k = kappa(labels,preds,weights="quadratic")
return 'error',k
def compute_kappa(data):
print "kappa: ",
print skll.kappa(data[0], data[1])
#print skll.kappa(data[0], data[1], weights='linear')
print "quadratic weighted kappa: ",
print skll.kappa(data[0], data[1], weights='quadratic')
# Collect the data needed for the accuracy assessment.
xys = []
classes = []
with open(accuracy_fn) as fp:
reader = csv.reader(fp)
next(reader)
for row in reader:
xys.append([float(n) for n in row[:2]])
classes.append(int(row[2]))
ds = gdal.Open(prediction_fn)
pixel_trans = gdal.Transformer(ds, None, [])
offset, ok = pixel_trans.TransformPoints(True, xys)
cols, rows, z = zip(*offset)
data = ds.GetRasterBand(1).ReadAsArray()
sample = data[rows, cols]
del ds
# Compute kappa.
print('Kappa:', skll.kappa(classes, sample))
# Create the confusion matrix.
labels = np.unique(np.concatenate((classes, sample)))
matrix = metrics.confusion_matrix(classes, sample, labels)
# Add labels to the matrix and save it.
matrix = np.insert(matrix, 0, labels, 0)
matrix = np.insert(matrix, 0, np.insert(labels, 0, 0), 1)
np.savetxt(matrix_fn, matrix, fmt='%1.0f', delimiter=',')
def Kappa(y_true, y_pred, **kwargs):
from numpy import clip
from skll import kappa
return kappa(clip(y_true, 1, 8), clip(y_pred, 1, 8), **kwargs)
def compute_kappa(data):
print "kappa: ",
print skll.kappa(data[0], data[1])
#print skll.kappa(data[0], data[1], weights='linear')
print "quadratic weighted kappa: ",
print skll.kappa(data[0], data[1], weights='quadratic')
valid_df = pd.read_csv(valid_fn)
class_type = ['裸地', '建筑', '农田', '林地', '工业用地', '潮滩', '稻田', '芦苇草地', '盐田', '水域']
# 获取图像坐标信息（xoff, yoff）
xys = np.array(valid_df[['xoff', 'yoff']])
# 提取类别标签数据
y = np.array(valid_df['class'], dtype=int)
del valid_df
cols, rows = zip(*xys)
del xys
data = tif_ds.GetRasterBand(1).ReadAsArray()
y_hat = data[rows, cols]
print('Kappa:', skll.kappa(y, y_hat))
labels = np.unique(np.concatenate((y, y_hat)))
matrix = confusion_matrix(y, y_hat, labels)
print("\nConfusion matrix:\n{}\n".format(matrix))
scores_image = mglearn.tools.heatmap(matrix,
xlabel='Predicted label',
ylabel='True label',
xticklabels=labels,
yticklabels=labels,
cmap='viridis',
fmt='%d')
plt.title("RF_opt_Confusion matrix")
plt.gca().invert_yaxis()
report = classification_report(y,
