def cusum(other_args: List[str], stock: pd.DataFrame):
"""Cumulative sum algorithm (CUSUM) to detect abrupt changes in data
Parameters
----------
other_args : str
Command line arguments to be processed with argparse
ticker : str
Ticker of the stock
stock : pd.DataFrame
Stock data
"""
parser = argparse.ArgumentParser(
add_help=False,
prog="cusum",
description="""
Cumulative sum algorithm (CUSUM) to detect abrupt changes in data
""",
)
parser.add_argument(
"-t",
"--threshold",
dest="threshold",
type=float,
default=(max(stock["Adj Close"].values) -
min(stock["Adj Close"].values)) / 40,
help="threshold",
)
parser.add_argument(
"-d",
"--drift",
dest="drift",
type=float,
default=(max(stock["Adj Close"].values) -
min(stock["Adj Close"].values)) / 80,
help="drift",
)
try:
ns_parser = parse_known_args_and_warn(parser, other_args)
if not ns_parser:
return
stock = stock["Adj Close"]
detect_cusum(stock.values, ns_parser.threshold, ns_parser.drift, True,
True)
if gtff.USE_ION:
plt.ion()
plt.show()
print("")
except Exception as e:
print(e, "\n")
return
def display_cusum(df: pd.DataFrame, target: str, threshold: float,
drift: float):
"""Cumulative sum algorithm (CUSUM) to detect abrupt changes in data
Parameters
----------
df : pd.DataFrame
Dataframe
target : str
Column of data to look at
threshold : float
Threshold value
drift : float
Drift parameter
"""
detect_cusum(df[target].values, threshold, drift, True, True)
if gtff.USE_ION:
plt.ion()
plt.show()
print("")
def calculate_cusum(data, threshold, drift):
ta, tai, taf, amp = detect_cusum(data['Data'], threshold, drift, ending=True, show=False)
data["Cumulative Sum Algorithm"] = 0
start = False
for i in range(5, len(data["Cumulative Sum Algorithm"])):
if i in tai:
start = True
elif i in taf:
start = False
if start:
data["Cumulative Sum Algorithm"].at[i] = 1
plt.clf()
return "Cumulative Sum Algorithm"
def cu_changepoint(f, window=25):
# change point detection based on cusum
# no generic way to set up threshold and drift
# f, window = pre_changepoint(ts_, window=window)
# ycol = 'dtrend_diff' # dtrend # used to detect changes in trend
# ts = f[ycol].values
ts = f['dtrend_diff'].values
if f['period'].isnull().sum() == 0:
window = f.loc[f.index[0], 'period']
min_ix = f.index.min()
m = 8.0
threshold = m * f['dtrend_diff'].std()
thres = threshold
while thres >= threshold / m:
drift = f['dtrend_diff'].std()
while drift > f['dtrend_diff'].std() / 10:
ta, tai, taf, amp = detect_cusum(ts, threshold=thres, drift=drift, ending=False, show=False)
print('thres: ' + str(thres) + ' drift: ' + str(drift) + ' ta: ' + str(min_ix + ta) + ' tai: ' + str(min_ix + tai) + ' taf: ' + str(taf))
drift /= 2
thres /= 2
def analyseStocks(symbol, investAmount, thresholdMargin=50000):
df = getDailyDataFromMarketstack(symbol, '2020-07-01')
if df.empty:
return ('empty', symbol, 0, 0)
minClose = min(df['Close'])
maxClose = max(df['Close'])
lastClose = df['Close'][-1]
difMinMax = maxClose - minClose
dateValue = pd.DataFrame()
dateValue['Date'] = df['Date']
dateValue['Value'] = df['Close']
lastFallDelta = getLastFallDelta(dateValue)
#logging.debug("minClose: ", minClose, " maxClose: ", maxClose, "lastClose: ", lastClose, " diffMinMax: ", difMinMax, ' lastFallDelta: ', lastFallDelta)
ta, tai, taf, amp = detect_cusum(df['Close'], 200, .05, True, False)
#logging.debug(ta, tai, taf, amp)
stocksToBuy = getAdjustedStocksToBuy(lastClose, investAmount)
totalInvestCost = lastClose * stocksToBuy + transactionTax(investAmount)
logging.debug("totalinvestcost: %s", totalInvestCost)
potentialStocksSold = maxClose * stocksToBuy + transactionTax(investAmount)
potentialRevenue = potentialStocksSold - totalInvestCost
logging.info("potentialRevenue last fall: %s", potentialRevenue)
if (int(potentialRevenue) > thresholdMargin):
logging.debug("Comprar")
return ('buy', symbol, lastClose, potentialRevenue)
return ('pass', symbol, lastClose, potentialRevenue)
def display_cusum(
df: pd.DataFrame,
target: str,
threshold: float,
drift: float,
external_axes: Optional[List[plt.Axes]] = None,
):
"""Cumulative sum algorithm (CUSUM) to detect abrupt changes in data
Parameters
----------
df : pd.DataFrame
Dataframe
target : str
Column of data to look at
threshold : float
Threshold value
drift : float
Drift parameter
external_axes : Optional[List[plt.Axes]], optional
External axes (2 axes are expected in the list), by default None
"""
target_series = df[target].values
# The code for this plot was adapted from detecta's sources because at the
# time of writing this detect_cusum had a bug related to external axes support.
# see https://github.com/demotu/detecta/pull/3
tap, tan = 0, 0
ta, tai, taf, _ = detect_cusum(
x=target_series,
threshold=threshold,
drift=drift,
ending=True,
show=False,
)
# Thus some variable names are left unchanged and unreadable...
gp, gn = np.zeros(target_series.size), np.zeros(target_series.size)
for i in range(1, target_series.size):
s = target_series[i] - target_series[i - 1]
gp[i] = gp[i - 1] + s - drift # cumulative sum for + change
gn[i] = gn[i - 1] - s - drift # cumulative sum for - change
if gp[i] < 0:
gp[i], tap = 0, i
if gn[i] < 0:
gn[i], tan = 0, i
if gp[i] > threshold or gn[i] > threshold: # change detected!
ta = np.append(ta, i) # alarm index
tai = np.append(tai, tap if gp[i] > threshold else tan) # start
gp[i], gn[i] = 0, 0 # reset alarm
if external_axes is None:
_, axes = plt.subplots(
2,
1,
sharex=True,
figsize=plot_autoscale(),
dpi=PLOT_DPI,
)
(ax1, ax2) = axes
else:
if len(external_axes) != 2:
logger.error("Expected list of two axis items.")
console.print("[red]Expected list of 2 axis items./n[/red]")
return
(ax1, ax2) = external_axes
target_series_indexes = range(df[target].size)
ax1.plot(target_series_indexes, target_series)
if len(ta):
ax1.plot(
tai,
target_series[tai],
">",
markerfacecolor=theme.up_color,
markersize=5,
label="Start",
)
ax1.plot(
taf,
target_series[taf],
"<",
markerfacecolor=theme.down_color,
markersize=5,
label="Ending",
)
ax1.plot(
ta,
target_series[ta],
"o",
markerfacecolor=theme.get_colors()[-1],
markeredgecolor=theme.get_colors()[-2],
markeredgewidth=1,
markersize=3,
label="Alarm",
)
ax1.legend()
ax1.set_xlim(-0.01 * target_series.size, target_series.size * 1.01 - 1)
ax1.set_ylabel("Amplitude")
ymin, ymax = (
target_series[np.isfinite(target_series)].min(),
target_series[np.isfinite(target_series)].max(),
)
y_range = ymax - ymin if ymax > ymin else 1
ax1.set_ylim(ymin - 0.1 * y_range, ymax + 0.1 * y_range)
ax1.set_title(
"Time series and detected changes " +
f"(threshold= {threshold:.3g}, drift= {drift:.3g}): N changes = {len(tai)}",
fontsize=10,
)
theme.style_primary_axis(ax1)
ax2.plot(target_series_indexes, gp, label="+")
ax2.plot(target_series_indexes, gn, label="-")
ax2.set_xlim(-0.01 * target_series.size, target_series.size * 1.01 - 1)
ax2.set_xlabel("Data points")
ax2.set_ylim(-0.01 * threshold, 1.1 * threshold)
ax2.axhline(threshold)
theme.style_primary_axis(ax2)
ax2.set_title(
"Time series of the cumulative sums of positive and negative changes",
fontsize=10,
)
ax2.legend()
if external_axes is None:
theme.visualize_output()
plt.plot(y)
plt.xlabel('Samples')
plt.show()'''
'''x1 = [0, 0, 3, 7, 9]
y1 = [0, 8, 12, 13, 14]
x2 = [0, 0, 1, 1, 2, 3, 9]
y2 = [0, 13, 13, 14, 14, 15, 15]
plt.xlim((0,9))
plt.ylim((0,16))
plt.xlabel('False Positive')
plt.ylabel('True Positive')
plt.plot(x1, y1, label = 'CUSUM')
plt.plot(x2, y2, label = 'NLMS')
plt.legend()
plt.show()'''
start2 = time()
ta, tai, taf, amp = detect_cusum(data_test, 2000, 10000, True, True)
end2 = time()
time1 = end1 - start1
time2 = end2 - start2
x1 = ['NLMS', 'CUSUM']
y1 = [time1, time2]
plt.bar(x1, y1)
plt.ylabel("Time(sec)")
plt.show()
a = 1
