def randomForest(request):
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
# target是补全的目标列名
target = request.POST['target']
# ref是用来生成拟合值的相关项列表,格式是以逗号将各项隔开的字符串，如：SibSp,Pclass,Fare,Parch
ref_str = request.POST['ref']
ref = ref_str.split(',')
ref.insert(0,target)#将target列名插入在ref最前面
print(ref)
target_df=data_train[ref]#将这些列的数据都取出来
#将数据分成已知目标项值和未知目标项值两部分
known_data= target_df[target_df[target].notnull()].as_matrix()
unknown_data = target_df[target_df[target].isnull()].as_matrix()
# y即目标值
y = known_data[:, 0]
# X即特征属性值
X = known_data[:, 1:]
# fit到RandomForestRegressor之中
rfr = RandomForestRegressor(random_state=0, n_estimators=2000, n_jobs=-1)
rfr.fit(X, y)
# 用得到的模型进行未知目标值结果预测
predictedAges = rfr.predict(unknown_data[:, 1::])
# 用得到的预测结果填补原缺失数据
data_train.loc[(data_train[target].isnull()), target] = predictedAges
return response_util.csv_info(data_train)
def scale_(request):
if "POST" == request.method:
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
#尺度变换目标列名
target_str = request.POST['target']
target = target_str.split(',')
# 尺度变换方法：log2、log10、ln、abs、sqrt
scale = request.POST['scale']
for each in target:
target_df = data_train[each]
if scale == 'log2':
data_train[each] = np.log2(target_df)
elif scale == 'log10':
data_train[each] = np.log10(target_df)
elif scale == 'ln':
data_train[each] = np.log(target_df)
elif scale == 'abs':
data_train[each] = np.abs(target_df)
elif scale == 'sqrt':
data_train[each] = np.sqrt(target_df)
else:
return response_util.wrong_info(
'输入的方法不包含在log2、log10、ln、abs、sqrt里')
return response_util.csv_info(data_train)
def setId(request):
if "POST" == request.method:
# 获取数据
f = request.FILES.get("csv_file")
filename=csv_util.upload(f)
data_train=pd.read_csv(filename)
os.remove(filename)
# 增加id列
count = len(data_train)
list = []
for num in range(0, count):
list.append(num)
data_train.insert(0, 'id', list)
return response_util.csv_info(data_train)
def standard(request):
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
# target是标准化的目标列名
target_str = request.POST['target']
target = target_str.split(',')
scaler = preprocessing.StandardScaler()
for each in target:
standard_data = scaler.fit_transform(data_train[each].values.reshape(-1,1))
data_train.drop([each], axis=1, inplace=True)
data_train[each] = standard_data
return response_util.csv_info(data_train)
def dummy(request):
if "POST" == request.method:
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
# target_str为需要因子化的列名,以逗号隔开，例：cabin,sex,pclass
target_str = request.POST['target']
target = target_str.split(',')
for each in target:
dummies = pd.get_dummies(data_train[each], prefix=each)
data_train = pd.concat([data_train, dummies], axis=1)
data_train.drop([each], axis=1, inplace=True)
return response_util.csv_info(data_train)
def soften(request):
if "POST" == request.method:
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
#平滑目标列名
target = request.POST['target']
# 平滑方法:百分位 per 、阈值 thresh
soften_method = request.POST['soften_method']
# 上下数据点（上下百分位或最大最小阈值）
min_value = int(request.POST['min'])
max_value = int(request.POST['max'])
# 取出目标数据
target_df = data_train[target].values.tolist()
if soften_method == 'per':
# 百分位平滑
if min_value < 0:
return response_util.wrong_info('百分位平滑，min值应不小于0')
elif max_value > 100:
return response_util.wrong_info('百分位平滑，max值应不大于100')
else:
min_value /= 100
max_value /= 100
# 列表中最小最大值
data_min = min(target_df)
data_max = max(target_df)
# 计算出来的范围
min_num = data_min + min_value * (data_max - data_min)
max_num = data_min + max_value * (data_max - data_min)
for i in range(0, len(target_df)):
if target_df[i] < min_num:
target_df[i] = min_num
elif target_df[i] > max_num:
target_df[i] = max_num
data_train[target] = target_df
elif soften_method == 'thresh':
# 阈值平滑
for i in range(0, len(target_df)):
if target_df[i] < min_value:
target_df[i] = min_value
elif target_df[i] > max_value:
target_df[i] = max_value
data_train[target] = target_df
else:
return response_util.wrong_info('输入的方法不包含在per/thresh里')
return response_util.csv_info(data_train)
def normalize(request):
if "POST" == request.method:
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
#归一化目标列名
target = request.POST['target']
# 取出目标列数据
mm = preprocessing.MinMaxScaler() # 归一化
target_str = request.POST['target']
target = target_str.split(',')
for each in target:
mm_data = mm.fit_transform(data_train[each].values.reshape(
-1, 1)) # 处理数据
data_train.drop([each], axis=1, inplace=True)
data_train[each] = mm_data
return response_util.csv_info(data_train)
def discrete(request):
if "POST" == request.method:
# 获取数据
f = request.FILES.get("csv_file")
filename = csv_util.upload(f)
data_train = pd.read_csv(filename)
os.remove(filename)
#离散化目标列名
target = request.POST['target']
# 离散化方法:等频 frequency 、等距 metric、聚类 cluster
discrete_method = request.POST['discrete_method']
# 离散区间
num = int(request.POST['num'])
target_df = data_train[target]
if discrete_method == 'metric':
# 等距离散化
data_train[target] = pd.cut(target_df, num, labels=range(num))
elif discrete_method == 'frequency':
# 等频率离散化
w = [1.0 * i / num for i in range(num + 1)]
w = target_df.describe(percentiles=w)[4:4 + num + 1]
w[0] = w[0] * (1 - 1e-10)
data_train[target] = pd.cut(target_df, w, labels=range(num))
elif discrete_method == 'cluster':
#基于聚类的离散化
kmodel = KMeans(n_clusters=num, n_jobs=4) # n_jobs是并行数，一般等于CPU数
kmodel.fit(target_df.values.reshape((len(target_df), 1)))
c = pd.DataFrame(kmodel.cluster_centers_).sort_values(0)
# rolling_mean表示移动平均，即用当前值和前2个数值取平均数，
# 由于通过移动平均，会使得第一个数变为空值，因此需要使用.iloc[1:]过滤掉空值。
w = c.rolling(2).mean().iloc[1:]
w = [0] + list(w[0]) + [target_df.max()
] # 把首末边界点加上，首边界为0，末边界为data的最大值120000
data_train[target] = pd.cut(
target_df, w, labels=range(num)) # cut函数实现将data中的数据按照w的边界分类。
else:
return response_util.wrong_info(
'输入的方法不包含在frequency/metric/cluster里')
return response_util.csv_info(data_train)