def test_original_test(NoTrendData, TrendData, arbitrary_1d_data):
# check with no trend data
NoTrendRes = mk.original_test(NoTrendData)
assert NoTrendRes.trend == 'no trend'
assert NoTrendRes.h == False
assert NoTrendRes.p == 1.0
assert NoTrendRes.z == 0
assert NoTrendRes.Tau == 0.0
assert NoTrendRes.s == 0.0
assert NoTrendRes.var_s == 0.0
assert NoTrendRes.slope == 0.0
# check with trendy data
TrendRes = mk.original_test(TrendData)
assert TrendRes.trend == 'increasing'
assert TrendRes.h == True
assert TrendRes.p == 0.0
assert TrendRes.Tau == 1.0
assert TrendRes.s == 64620.0
np.testing.assert_allclose(TrendRes.slope, 1.0, rtol=1e-02)
# check with arbitrary data
result = mk.original_test(arbitrary_1d_data)
assert result.trend == 'no trend'
assert result.h == False
assert result.p == 0.37591058740506833
assert result.z == -0.8854562842589916
assert result.Tau == -0.03153167653875869
assert result.s == -1959.0
assert result.var_s == 4889800.333333333
assert result.slope == -0.0064516129032258064
def test_residuals(model, timeperiod, reg):
"""
Test for a residual trend, applying a Mann-Kendall-test
Parameters
----------
model : GLMObject
Best model
timeperiod : np.array
considered years (not used here)
Returns
-------
float
slope in residuals
float
p-value
"""
res_trend = mk.original_test(model.resid_response, alpha=0.1)
fig, ax = plt.subplots(figsize=(12, 8))
sm.graphics.tsa.plot_acf(model.resid_response, lags=39, ax=ax)
ax.set_xlabel('lag')
ax.set_title('Autocorrelation {}'.format(reg))
#fig.savefig('/home/insauer/projects/NC_Submission/Climada_papers/Test/AutocorrResidualsGMT_{}.png'.format(reg),bbox_inches = 'tight',dpi =600)
alt_trend_test = mk.hamed_rao_modification_test(model.resid_response)
return res_trend.slope, res_trend.p, alt_trend_test.trend, alt_trend_test.p
def mk_col_slope(Data, YearCol, LocCol, window, TH, alpha=0.1):
length = Data[LocCol].shape[0]
start_index = Data[YearCol].values[window - 1]
final_index = Data[YearCol].values[length - 1] + 1
Year = pd.Series(range(start_index, final_index))
iterations = length - (window - 1)
stats_list2 = pd.DataFrame(index=range(0, iterations),
columns=['Year', LocCol])
for instance in range(0, iterations):
stats_list2['Year'].values[instance] = Year[instance]
for instance in range(0, iterations):
snip = Data[LocCol].loc[instance:instance + window - 1]
if missing_values(snip, window, TH):
snip_test = MK.original_test(snip, alpha)
stats_list2[LocCol].values[instance] = snip_test.slope
else:
stats_list2[LocCol].values[instance] = 'Not enough data'
#print(stats_list2)
return stats_list2
def main():
parser = argparse.ArgumentParser()
parser.add_argument(
'--granularity',
'-g',
type=str,
default='minute',
help=
'one of [day,hour,minute], the granularity of the x-axis. default=minute'
)
parser.add_argument(
'--num_points',
'-n',
type=int,
default=60 * 24,
help=
'number of points to sample from historic prices (e.g. number of days). default=1440'
)
parser.add_argument(
'--mode',
'-m',
type=str,
default='close',
help=
'the cryptocompare price attribute to use (e.g. open or close). default=close'
)
parser.add_argument('--currency',
'-c',
type=str,
default='CAD',
help='currency (e.g. CAD, USD). default=CAD')
args = parser.parse_args()
fig, axs = plt.subplots(2, 1)
colors = 'rgbykc'
for ax, symbol, color in zip(axs, ['BTC', 'ETH'], colors):
fn = getattr(cryptocompare, f'get_historical_price_{args.granularity}')
btc = fn(symbol, 'CAD', limit=args.num_points)
data = [x[args.mode] for x in btc]
cur = data[-1]
result = mk.original_test(data)
line = lambda t: result.slope * t + result.intercept
print(symbol, ':', result, '\n\n')
ax.plot(
data,
label=f'{symbol} = {cur} ({result.trend}, p={round(result.p, 4)})',
c=color)
ax.plot(list(map(line, range(len(data)))), label='MK fit')
ax.legend()
ax.set_ylabel(f'{symbol} Price ({args.mode})')
ax.set_xlabel('previous ' + args.granularity + 's')
fig.suptitle(
f'Crypto prices {args.currency} for the last {args.num_points} {args.granularity}s',
fontsize=12)
fig.tight_layout()
plt.show()
def multip_function(x):
"""This function estimates the trend in the discharge time series of one grid cell.
Parameters
----------
x : tuple
lat, lon coordinates
Returns
-------
reg.slope
slope of the trend
reg.p
v-value of the trend
"""
lat, lon = x
# exclude coordinates over sea
if (lat == -1000) or (lon == -1000):
return np.nan, np.nan
print(lon, lat)
data = get_dis_gridcell(lat, lon)
if np.isnan(data).all():
return np.nan, np.nan
else:
reg = mk.original_test(data, alpha=0.1)
return reg.slope, reg.p
def test_crash(norm_stats, crash_date, norm_name):
(V, SD, AC) = norm_stats
sys.stdout.write(f"Results of the Mann Kendall Test "
f"for the {norm_name} (crash: {crash_date}): \n")
MKV = mk.original_test(V)
sys.stdout.write(
f"Variance: trend = {MKV.trend} | tau = {MKV.Tau:0.4f}\n")
MKSD = mk.original_test(SD)
sys.stdout.write(
f"Spectral Density: trend = {MKSD.trend} | tau = {MKSD.Tau:0.4f}\n"
)
MKAC = mk.original_test(AC)
sys.stdout.write(
f"Autocorrelation: trend = {MKAC.trend} | tau = {MKAC.Tau:0.4f}\n\n"
)
def test_residuals(model, timeperiod, reg):
"""
Test for a residual trend, applying a Mann-Kendall-test
Parameters
----------
model : GLMObject
Best model
timeperiod : np.array
considered years (not used here)
Returns
-------
float
slope in residuals
float
p-value
"""
res_trend = mk.original_test(model.resid_response, alpha=0.1)
fig, ax = plt.subplots(figsize=(12, 8))
alt_trend_test = mk.hamed_rao_modification_test(model.resid_response)
return res_trend.slope, res_trend.p, alt_trend_test.trend, alt_trend_test.p
def mann_kendall_price(self):
"""
The Mann Kendall Trend Test (sometimes called the M-K test) is used to analyze data collected over time for
consistently increasing or decreasing trends (monotonic) in Y values. H0 - there is no monotonic trend,
H1 - the trend exists, it is either positive or negative
:return: trend: trend direction, h: bool if trend exists, p: p-value, z: z-stat, Tau: Kendall Tau,
s: Mann-Kendal’s score, var_s: Variance S, slope: Theil-Sen estimator/slope,
intercept: Intercept of Kendall-Theil Robust Line
"""
return mk.original_test(self.center_price())
def doTheAnnMeanFlowTrends(pastStats, currentStats = None):
tSeries = []
for key in pastStats:
#tSeries.extend(pastStats[key][:12])
tSeries.append(pastStats[key][12])
#print(pastStats[key])
if currentStats is not None:
for key in currentStats:
tSeries.append(currentStats[key][12])
#print(len(tSeries), sum(tSeries)/len(tSeries))
#print(tSeries)
return [pmk.original_test(tSeries), pmk.hamed_rao_modification_test(tSeries)]
def mkTest(series, seasonal):
if seasonal == False:
data_mk = mk.original_test(series)
trend = data_mk[0]
else:
data_mk_seasonal_test = mk.seasonal_test(series, period= 12)
trend = data_mk_seasonal_test[0]
if trend == 'decreasing' or trend == 'increasing':
self.__trend__ = 'present'
trend = 'present'
return trend
self.__trend__ = trend
return trend
def trendTest(output_dir, data):
"""
Tests each of the call categories during a given time period for
a monotonic trend by using the mann-kendall test. Results of this
test are saved to a csv file.
Inputs:
- output_dir: String path to output directory
- data: incident data of time period you want to graph
"""
call_categories = [
'injuries_external', 'motor', 'health', 'fire', 'mental_illness',
'other'
]
trend = []
h = []
p = []
z = []
Tau = []
s = []
var_s = []
slope = []
for category in call_categories:
category_data = data[[category]]
to, ho, po, zo, Tauo, so, var_so, slopeo = mk.original_test(
category_data)
trend.append(to)
h.append(ho)
p.append(po)
z.append(zo)
Tau.append(Tauo)
s.append(so)
var_s.append(var_so)
slope.append(slopeo)
results = pd.DataFrame({
'Call_Category': call_categories,
'Trend': trend,
'h': h,
'p': p,
'z': z,
'Tau': Tau,
's': s,
'var_s': var_s,
'slope': slope
})
results.to_csv(join(output_dir, "trend_test_results.csv"), index=False)
def getValuesMK(arry, p_sig):
z_, lines, cols = arry.shape
arryMk = np.ndarray( ( 3, lines, cols ) )
for l in range(lines):
for c in range(cols):
v = arry[:, l, c ]
if np.isnan( np.sum(v) ):
arryMk[0, l, c] = np.nan
arryMk[1, l, c] = np.nan
arryMk[2, l, c] = np.nan
continue
r = mk.original_test( v, p_sig )
arryMk[0, l, c] = r.s if r.p <= p_sig else np.nan
arryMk[1, l, c] = r.p
arryMk[2, l, c] = r.slope if r.p <= p_sig else np.nan
return arryMk
def test_autocorrelation(time_series):
"""
Test for autocorrelation
Parameters
----------
time-series
Returns
-------
float
tau
"""
auto = mk.original_test(time_series, alpha=0.1)
return auto.Tau
def determine_orientation(self, positions):
"Given a list of minimizer positions, determine the orientation of the contig"
if len(positions) > 1:
if all(x < y for x, y in zip(positions, positions[1:])):
return "+"
if all(x > y for x, y in zip(positions, positions[1:])):
return "-"
if self.args.mkt:
mkt_result = mk.original_test(positions)
if mkt_result.h and mkt_result.p <= 0.05:
return "+" if mkt_result.trend == "increasing" else "-"
else:
tally = Counter([x < y for x, y in zip(positions, positions[1:])])
positive_perc = tally[True]/float(len(positions)-1)*100
negative_perc = 100 - positive_perc
if positive_perc >= self.args.m:
return "+"
if negative_perc >= self.args.m:
return "-"
return "?"
def mk_column(Data, YearCol, LocCol, windows, datalist, threshold, alpha=0.1):
for each in windows:
length = Data[LocCol].shape[0]
start_index = Data[YearCol].values[each - 1]
final_index = Data[YearCol].values[length - 1] + 1
Year = pd.Series(range(start_index, final_index))
iterations = length - (each - 1)
stats_list2 = pd.DataFrame(index=range(0, iterations),
columns=[
LocCol, 'Year', 'trend', 'Ha', 'p', 'Z',
'S', 'VAR(S)', 'slope'
])
stats_list2[LocCol].values[0] = str(each) + " Year window"
for instance in range(0, iterations):
stats_list2['Year'].values[instance] = Year[instance]
for instance in range(0, iterations):
snip = Data[LocCol].loc[instance:instance + each -
1] # INSERT TEST FOR 10%
if missing_values(snip, each, threshold):
snip_test = MK.original_test(snip, alpha)
stats_list2['trend'].values[instance] = snip_test.trend
stats_list2['Ha'].values[instance] = snip_test.h
stats_list2['p'].values[instance] = snip_test.p
stats_list2['Z'].values[instance] = snip_test.z
stats_list2['S'].values[instance] = snip_test.s
stats_list2['VAR(S)'].values[instance] = snip_test.var_s
stats_list2['slope'].values[instance] = snip_test.slope
else:
stats_list2['trend'].values[instance] = 'Not enough data'
#print(stats_list2)
datalist.append(
stats_list2
) # this is used if you want to enter different data frames in a list
''' need to find way to store it in Excel '''
def test_autocorrelation(time_series):
"""
Test for a residual trend, applying a Mann-Kendall-test
Parameters
----------
model : GLMObject
Best model
timeperiod : np.array
considered years (not used here)
Returns
-------
float
slope in residuals
float
p-value
"""
auto = mk.original_test(time_series, alpha=0.1)
return auto.Tau
def test_residuals(model, timeperiod, reg):
"""
Test for a residual trend, applying a Mann-Kendall-test
Parameters
----------
model : GLMObject
Best model
timeperiod : np.array
considered years (not used here)
Returns
-------
float
slope in residuals
float
p-value
"""
res_trend = mk.original_test(model.resid_response, alpha=0.1)
return res_trend.slope, res_trend.p
def calc_mann_kendall(data_file, info_file, out_file):
"""
用 kendall tau刻画每个column的发育趋势
"""
# load
df = pd.read_csv(data_file)
info_df = pd.read_csv(info_file)
ages = np.array(info_df['age in years'])
age_uniq = np.unique(ages)
# calculate
out_df = pd.DataFrame(index=('tau', 'p'), columns=df.columns)
for col in out_df.columns:
meas_vec = np.array(df[col])
y = np.zeros_like(age_uniq, dtype=np.float64)
for age_idx, age in enumerate(age_uniq):
y[age_idx] = np.mean(meas_vec[ages == age])
mk_test = mk.original_test(y, 0.05)
out_df.loc['tau', col] = mk_test.Tau
out_df.loc['p', col] = mk_test.p
# save
out_df.to_csv(out_file)
def Trend_1(data_sku, star1):
starting = star1
data_sku3 = data_sku[starting:len(data_sku)]
data_sku3 = data_sku3.reset_index(drop=True)
index = data_sku3.ne(0).idxmax()
data_sku4 = data_sku3[index:] #first non zero element
data_sku4 = data_sku4.reset_index(drop=True)
Zero = np.where((data_sku4 == 0) == True)
sparsity = len(Zero) / len(data_sku4)
if len(data_sku3) >= 8 and sparsity < (0.2):
dd = mk.original_test(data_sku3)
if dd[0] == 'increasing':
Type_0 = 'Growing'
elif dd[0] == 'decreasing':
Type_0 = 'Degrowing'
else:
Type_0 = 'Normal'
elif len(data_sku3) >= 12 and sparsity > (0.2):
#rolling-->can be applied to series
roll_sum = pd.Series(data_sku).rolling(6).apply(np.mean)
roll_diff = np.diff(roll_sum)
g_1 = len(np.where((roll_diff > 0) == True)[0])
l_1 = len(np.where((roll_diff < 0) == True)[0])
g_p = g_1 / len(roll_diff)
l_p = l_1 / len(roll_diff)
if g_p >= 0.75:
Type_0 = 'Growing'
elif l_p >= 0.75:
Type_0 = 'Degrowing'
else:
Type_0 = 'Normal'
else:
Type_0 = 'Normal'
# v=c(Type_0,starting)
return (Type_0)
def mk_column(Data, YearCol, LocCol, windows):
for each in windows:
length = Data[LocCol].shape[0]
start_index = Data[YearCol].values[each - 1]
final_index = Data[YearCol].values[length - 1] + 1
Year = pd.Series(range(start_index, final_index))
iterations = length - (each - 1)
stats_list2 = pd.DataFrame(
index=range(0, iterations),
columns=['Year', 'trend', 'Ha', 'p', 'Z', 'S', 'VAR(S)', 'slope'])
stats_list2.insert(0, 'Name', None)
stats_list2['Name'].values[0] = LocCol
for instance in range(0, iterations):
stats_list2['Year'].values[instance] = Year[instance]
for instance in range(0, iterations):
snip = Data[LocCol].loc[instance:instance + each - 1]
snip_test = MK.original_test(snip, 0.1)
stats_list2['trend'].values[instance] = snip_test[0]
stats_list2['Ha'].values[instance] = snip_test[1]
stats_list2['p'].values[instance] = snip_test[2]
stats_list2['Z'].values[instance] = snip_test[3]
stats_list2['S'].values[instance] = snip_test[5]
stats_list2['VAR(S)'].values[instance] = snip_test[6]
stats_list2['slope'].values[instance] = snip_test[7]
# stats_list2 = stats_list2.transpose()
# stats_list2.insert(stats_list2.shape[1],None,None)
# stats_list2 = stats_list2.transpose()
stats_list2 = stats_list2.append(pd.Series([np.nan]),
ignore_index=True)
print(stats_list2)
# stats_list2.to_excel("testNewColTranspose.xlsx", index=False)
''' need to find way to store it in Excel '''
def add_element(self, value):
'''
Add new element to the statistic
'''
#reset parameters if change was detected:
if self.in_concept_change:
self.reset()
#append elements:
self.instance_memory.append(value)
if len(self.instance_memory) == self.min_instances:
self.sample_count = 1
if len(self.instance_memory) > self.min_instances:
self.instance_count += 1
#start drift detection: >> min_instances have to be reached, then always perform test once, after that perform test every i_th instance (instances_step)
if len(self.instance_memory) >= self.min_instances and ((self.instance_count == self.instances_step) or (self.sample_count == 1)):
if self.test_type == 'original_mk':
#call corresponding test from package:
print('Perform MK test')
results_tuple = mk.original_test(self.instance_memory, self.alpha)
print('MK test ended')
if self.test_type == 'hamed_rao_mod':
#call corresponding test from package:
results_tuple = mk.hamed_rao_modification_test(self.instance_memory, self.alpha)
if self.test_type == 'yue_wang_mod':
#call corresponding test from package:
results_tuple = mk.yue_wang_modification_test(self.instance_memory, self.alpha)
if self.test_type == 'trend_free_pre_whitening_mod':
#call corresponding test from package:
results_tuple = mk.trend_free_pre_whitening_modification_test(self.instance_memory, self.alpha)
if self.test_type == 'pre_whitening_mod':
#call corresponding test from package:
results_tuple = mk.pre_whitening_modification_test(self.instance_memory, self.alpha)
if self.test_type == 'seasonal':
#call corresponding test from package:
results_tuple = mk.seasonal_test(self.instance_memory, period = self.period, alpha = self.alpha)
#reset counter every time a test was performed:
self.sample_count = 0
self.instance_count = 0
#assign results:
self.p_value = results_tuple[2]
self.sens_slope = results_tuple[-1]
self.trend = results_tuple[0]
if self.p_value < self.alpha and np.abs(self.sens_slope) > self.slope_threshold:
self.in_concept_change = True
else:
self.in_concept_change = False
def rel_time_attr_MK(dataFrame71):
"""
Theil-Sen-Slope estimation and Mann-Kendall-Test to estimate the
contribution of each driver!
Parameters
----------
dataFrame71 : time series
Time series
Returns
-------
regH : List MK-output
Sen_slope and MK-test result with uncertainty range of hazard
(with 1980 fixed exposure)(TS_Haz) 1980-2010
regHE : List MK-output
Sen_slope and MK-test result with uncertainty range of TS_HazExp
1980-2010
regF : List MK-output
Sen_slope and MK-test result with uncertainty range of TS_Full
1980-2010.
regH7 : List MK-output
Sen_slope and MK-test result with uncertainty range of hazard
(with 1980 fixed exposure)(TS_Haz) 1971-2010
regH107 : List MK-output
Sen_slope and MK-test result with uncertainty range of hazard
(with 2010 fixed exposure)(TS_Haz) 1971-2010
regH10 : List MK-output
Sen_slope and MK-test result with uncertainty range of hazard
(with 2010 fixed exposure)(TS_Haz) 1980-2010
regE : List MK-output
Sen_slope and MK-test result with uncertainty range of exposure
difference function (TS_HazExp - TS_Haz) 1980-2010 (not used)
regE7 : List MK-output
Sen_slope and MK-test result with uncertainty range of exposure
difference function (TS_HazExp - TS_Haz) 1971-2010 (not used)
regV : List MK-output
Sen_slope and MK-test result with uncertainty range of vulnerability
difference function (TS_full - TS_Haz_Exp)(not used)
regI : List MK-output
Sen_slope and MK-test result with uncertainty range of modeled damges
(including vulnerability)
regN : List MK-output
Sen_slope and MK-test result with uncertainty range of observed damages
"""
dataFrame = dataFrame71[dataFrame71['Year'] > 1979]
regLHazExp = mk.original_test(dataFrame['Norm_Impact_2y_trend'], alpha=0.1)
slopeLHazExp = stats.theilslopes(dataFrame['Norm_Impact_2y_trend'],
alpha=0.1)
regHE = [regLHazExp.slope, regLHazExp.p, slopeLHazExp[2], slopeLHazExp[3]]
regLFull = mk.original_test(dataFrame['Norm_Impact_Pred'], alpha=0.1)
slopeLFull = stats.theilslopes(dataFrame['Norm_Impact_Pred'], alpha=0.1)
regF = [regLFull.slope, regLFull.p, slopeLFull[2], slopeLFull[3]]
regHaz = mk.original_test(dataFrame['Norm_ImpFix_2y_trend'], alpha=0.1)
slopeHaz = stats.theilslopes(dataFrame['Norm_ImpFix_2y_trend'], alpha=0.1)
regH = [regHaz.slope, regHaz.p, slopeHaz[2], slopeHaz[3]]
regHaz7 = mk.original_test(dataFrame71['Norm_ImpFix_2y_trend'], alpha=0.1)
slopeHaz7 = stats.theilslopes(dataFrame71['Norm_ImpFix_2y_trend'],
alpha=0.1)
regH7 = [regHaz7.slope, regHaz7.p, slopeHaz7[2], slopeHaz7[3]]
regHaz107 = mk.original_test(dataFrame71['Norm_Imp2010_2y_trend'],
alpha=0.1)
slopeHaz107 = stats.theilslopes(dataFrame71['Norm_Imp2010_2y_trend'],
alpha=0.1)
regH107 = [regHaz107.slope, regHaz107.p, slopeHaz107[2], slopeHaz107[3]]
regHaz10 = mk.original_test(dataFrame['Norm_Imp2010_2y_trend'], alpha=0.1)
slopeHaz10 = stats.theilslopes(dataFrame['Norm_Imp2010_2y_trend'],
alpha=0.1)
regH10 = [regHaz10.slope, regHaz10.p, slopeHaz10[2], slopeHaz10[3]]
regNat = mk.original_test(dataFrame['natcat_flood_damages_2005_CPI'],
alpha=0.1)
slopeNat = stats.theilslopes(dataFrame['natcat_flood_damages_2005_CPI'],
alpha=0.1)
regN = [regNat.slope, regNat.p, slopeNat[2], slopeNat[3]]
return regH, regHE, regH7, regH107, regH10, regF, regN
SLS = Data_2_use.sel(lon=0.5, lat=7.5, method='nearest')
SLS.plot()
#Areal Selection & Averaging
ASA = Data_2_use.sel(lon=np.arange(-1.5, 1.5, 0.5),
lat=np.arange(5, 15, 0.5),
method='nearest')
ASA = ASA.mean(dim=('lon', 'lat'))
ASA.plot()
#Annual Averaging
Ann_avg = ASA.groupby('time.year').mean('time')
Ann_avg.plot()
### Statistics of Data
stat_result = pmk.original_test(Ann_avg)
print(stat_result)
#Seasonal Climatology
Seas_avg = ASA.groupby('time.season').mean('time')
print(Seas_avg)
#THANK YOU
#Reading Data from NetCDF File
Data = xr.open_dataset(in_file)
Data_2_use = Data['tmp']
#Selecting location 0.5W, 6.5N
loc_data = Data_2_use.sel(lon=-0.5, lat=6.5, method='nearest')
loc_data.plot()
break
x = struct.unpack("f", x)[0]
if (not np.isinf(x) and not np.isnan(x)):
ajat.append(x)
ind.append(i)
else:
dnfind.append(i)
i += 1
f.close()
ajat = np.array(ajat)
dnfind = np.array(dnfind)
ind = np.array(ind)
sr = pd.Series(ajat, ind)
ts = mk.original_test(sr)
pns = stat.linregress(ind, ajat)
plt.plot(ind, ajat, 'o', color='b')
plt.plot(ind, ind * pns.slope + pns.intercept, label="pns")
plt.plot(ind, ind * ts.slope + ts.intercept, label="ts")
plt.xlim(right=np.max(ind))
plt.ylim(top=np.max(ajat))
#dnfien plottaus
ala, ula = plt.ylim()
dnfaika = np.zeros(np.shape(dnfind)) + ula
plt.plot(dnfind, dnfaika, 'o', color='r')
plt.legend()
#plt.tight_layout();
def trend_test(df):
result = mk.original_test(df)
print(result)
import matplotlib.pyplot as plt
import statsmodels.api as sm
from statsmodels.tsa.seasonal import seasonal_decompose
from statsmodels.tsa.api import Holt
from statsmodels.tsa.stattools import adfuller
import pymannkendall as mk
import os
#We imported the required libraries
filenamemadrid = os.getcwd() + "\\weather_madrid_LEMD_1997_2015.csv"
df_madrid = pd.read_csv(filenamemadrid,
usecols=["CET", "Mean TemperatureC"],
sep=",")
#We make python read the csv files
df_madrid = df_madrid.dropna()
df_madrid = df_madrid.rename(columns={"CET": "date"})
filenamebrazil = os.getcwd() + "\\sudeste.csv"
df_brazil = pd.read_csv(filenamebrazil, usecols=["date", "temp"], sep=",")
df_brazil = df_brazil.dropna()
df_brazil = df_brazil.groupby(["date"])["temp"].mean()
df_brazil = df_brazil.to_frame()
df_brazil = df_brazil.reset_index(drop=False)
df_brazil = df_brazil.set_index('date')
df_madrid = df_madrid.set_index('date')
plt.show()
trend_brazil = mk.original_test(df_brazil)
print(trend_brazil)
trend_madrid = mk.original_test(df_madrid)
print(trend_madrid)
def do_backtest(df, symbol, end=None):
trade_count = 0
trade_history = []
balance = initial_balance
win_count = 0
loss_count = 0
profit = 0
action = HOLD
current_tick = 0
entry_tick = 0
buy_mode = True
entry_price = 0
buy_index = 0
window_size = 1000
last_size = 50
if backtest_mode == 2:
df = df.iloc[end - window_size * 1 - 100:end + window_size * 2]
elif backtest_mode == 3:
df_x = df
df = df.iloc[195267:199267]
# fragment = detect_anomaly(df)
#detect_anomaly(df.iloc[11706:11074])
#plot_whole(df_x)
df = df.reset_index()
df = df.fillna(0)
for i, row in df.iterrows():
start_time = time.time()
current_price = row['last_price']
current_ask_price = row['best_ask_price']
current_bid_price = row['best_bid_price']
current_tick += 1
if i > window_size:
last = df.iloc[i, :]
prev1 = df.iloc[i - 2, :]
prev25 = df.iloc[i - 25, :]
prev50 = df.iloc[i - 50, :]
prev100 = df.iloc[i - 100, :]
prev200 = df.iloc[i - 200, :]
prev500 = df.iloc[i - 500, :]
diffx1 = last.qav_sma500 - last.qav_sma1000
diffx2 = prev50.qav_sma500 - prev50.qav_sma1000
diffx3 = prev100.qav_sma500 - prev100.qav_sma1000
diffx4 = prev200.qav_sma500 - prev200.qav_sma1000
first_check = (
last['last_sma600'] > prev100['last_sma600']
and last['last_sma600'] > prev500['last_sma600']
and last.qav_sma500 > last.qav_sma1000
and prev50.qav_sma500 > prev50.qav_sma1000
and prev100.qav_sma500 > prev100.qav_sma1000
and prev200.qav_sma500 > prev200.qav_sma1000
and last.qav_sma500 > prev50.qav_sma500 > prev100.qav_sma500 >
prev200.qav_sma500 and diffx1 > diffx2 > diffx3 > diffx4
and diffx1 > 0.3 and ###buda yanıltıcı!!!!!
diffx1 < 1 ###yanıltıcı!!!!!
)
# if last['index'] == 114395:
# pdb.set_trace()
if (first_check == True and conditions[0]['buy_mode'] == True):
fragment = df.iloc[i - window_size:i, :]
fragment = detect_anomaly(fragment)
fragment = fragment.reset_index()
last = fragment.iloc[-1, :]
prev1 = fragment.iloc[-2, :]
first_n = fragment[:window_size - last_size]
last_n = fragment[-last_size:]
mk_test = mk.original_test(fragment.change_qav.to_numpy())
fragment_sum = fragment.groupby(
['score_qav', 'label_qav'], as_index=False,
sort=False)[["change_qav", "change_price"]].sum()
conditions[0]['buy_cond'] = (
(fragment_sum[fragment_sum['label_qav'] == 1].change_qav <
3).all() and mk_test.z > 1 and mk_test.z < 10
and mk_test.Tau < 0.1 and fragment_sum[
fragment_sum['label_qav'] == 1].change_qav.sum() > 4
and fragment_sum[fragment_sum['label_qav'] ==
1].change_qav.sum() < 10
and fragment_sum.label_qav.iloc[0] == 0
and fragment_sum.label_qav.iloc[-1] == 1
and fragment_sum.label_qav.iloc[-2] == 1 and
(fragment_sum[fragment_sum['label_qav'] == 0].change_qav <
fragment_sum[fragment_sum['label_qav'] ==
1].change_qav.max()).all()
and fragment_sum.iloc[-1].change_price +
fragment_sum.iloc[-2].change_price > 0
and fragment_sum.change_price.sum() > 0
and (last_n.label_qav == 1).count() < 50)
elif (conditions[0]['buy_mode'] == False):
conditions[0]['sell_cond'] = (last['last_sma600'] <
prev1['last_sma600'])
else:
continue
for ic, cond in enumerate(conditions):
if cond['buy_mode'] and cond['buy_cond']:
conditions[ic]['action'] = BUY
conditions[ic]['entry_price'] = current_ask_price
conditions[ic]['buy_mode'] = False
if ic == 0:
printLog("CONDITION " + str(ic + 1) + " IS BUYING....")
printLog("##### TRADE " + str(cond['trade_count']) +
" #####")
printLog("BUY: " + symbol + " for " +
str(cond['entry_price']) + " at " +
str(last.date) + " - index: " +
str(last['index']))
printLog(fragment[[
'index', 'date', 'symbol', 'last_price',
'total_traded_quote_asset_volume', 'label_qav',
'score_qav', 'change_qav', 'change_price'
]].tail(100))
printLog(
mk.original_test(fragment.change_qav.to_numpy()))
printLog(fragment_sum)
printLog("diffx1: " + str(diffx1))
printLog("last.qav_sma500: " + str(last.qav_sma500))
printLog("last.qav_sma1000: " + str(last.qav_sma1000))
printLog("prev100.qav_sma500: " +
str(prev100.qav_sma500))
printLog("prev100.qav_sma1000: " +
str(prev100.qav_sma1000))
#plot_whole(df)
#pdb.set_trace()
elif not cond['buy_mode'] and cond['sell_cond']:
printLog("CONDITION " + str(ic + 1) + " IS SELLING....")
conditions[ic]['action'] = SELL
exit_price = current_bid_price
profit = (
(exit_price - cond['entry_price']) /
cond['entry_price'] + 1) * (1 - transaction_fee)**2 - 1
conditions[ic]['balance'] = conditions[ic]['balance'] * (
1.0 + profit)
conditions[ic]['trade_count'] += 1
conditions[ic]['buy_mode'] = True
printLog("SELL: " + symbol + " for " + str(exit_price) +
" at " + str(last.date) + " - index: " +
str(last['index']))
printLog("PROFIT: " + str(profit * 100))
printLog("BALANCE: " + str(cond['balance']))
else:
conditions[ic]['action'] = HOLD
if (current_tick > len(df) - 1):
printLog("*********TOTAL RESULTS*************************")
for ic, cond in enumerate(conditions):
printLog("SYMBOL: " + symbol)
printLog("CONDITION NUMBER: " + str(ic))
printLog("TOTAL BALANCE: " + str(cond['balance']))
printLog("TRADE COUNT: " + str(cond['trade_count']))
printLog("**********************************")
if i % 1000 == 0:
printLog(symbol + "-" + str(row['index']))
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import pymannkendall as mk
Birth_data = pd.read_csv("daily-total-female-births.csv",
parse_dates=['Date'],
index_col='Date') #Birth_data
data = Birth_data
fig, ax = plt.subplots(figsize=(12, 8))
res = mk.original_test(data)
trend_line = np.arange(len(data)) * res.slope + res.intercept
ax.plot(data)
ax.plot(data.index, trend_line)
ax.legend(['data', 'trend line'])
ax.set(xlabel="Dates", ylabel="Births", title="Trend line")
fig.savefig('Trendline_plot1.png')
# IMPORTING THE FIRST DATA SET
Birth_data = pd.read_csv("daily-total-female-births.csv",
parse_dates=['Date'],
index_col='Date')
# SUMMARY STATISTICS
head = Birth_data.head()
Summary = Birth_data.describe()
print(head)
print(Summary)
# MANNKENDALL TREND TEST
Birth_data = pd.read_csv("daily-total-female-births.csv",
parse_dates=['Date'],
index_col='Date') #Birth_data
MKT = mk.original_test(Birth_data, alpha=0.05)
print(MKT)
# IMPORTING SECOND DATA SET
Shampoo_data = pd.read_csv("shampoo.csv",
parse_dates=['Month'],
index_col='Month')
# SUMMARY STATISTICS
head_shampoo = Shampoo_data.head()
Summary_shampoo = Shampoo_data.describe()
print(head_shampoo)
print(Summary_shampoo)
# TREND TEST 1
MKT1 = mk.hamed_rao_modification_test(Shampoo_data)
for month in range(1, 13, 1):
name_month = datetime.date(1900, int(month), 1).strftime('%B')
TREND = []
for index, row in list_nom.iterrows():
data = pd.read_csv(path_m + row[0] + '_MONTH_' + varin + '_' +
indice + '_' + str(yearmin) + '_' +
str(yearmax) + '_' +
str('{:02d}'.format(month)) + '.csv',
skiprows=2)
data = data.rename(columns={
data.columns[1]: "var"
}).set_index('datetime')
if (valeur['name'] == 'Original Mann-Kendall test'):
trend, h, p, z, Tau, s, var_s, slope, intercept = mk.original_test(
data)
elif (valeur['name'] == 'Hamed and Rao Modified MK Test'):
trend, h, p, z, Tau, s, var_s, slope, intercept = mk.hamed_rao_modification_test(
data)
elif (valeur['name'] == 'Yue and Wang Modified MK Test'):
trend, h, p, z, Tau, s, var_s, slope, intercept = mk.yue_wang_modification_test(
data)
elif (valeur['name'] ==
'Modified MK test using Pre-Whitening method'):
trend, h, p, z, Tau, s, var_s, slope, intercept = mk.pre_whitening_modification_test(
data)
elif (valeur['name'] ==
'Modified MK test using Trend free Pre-Whitening method'):
trend, h, p, z, Tau, s, var_s, slope, intercept = mk.trend_free_pre_whitening_modification_test(
data)
