-
Bert Palm authored677769a7
core.py 9.28 KiB
#! /usr/bin/env python
# -*- coding: utf-8 -*-
import numpy as np
import pandas as pd
import matplotlib as mpl
from config import Fields, Params
from funcs import flagDispatch
from dsl import parseFlag
from flagger import PositionalFlagger, BaseFlagger
def inferFrequency(data):
return pd.tseries.frequencies.to_offset(pd.infer_freq(data.index))
def flagWindow(flagger, flags, mask, direction='fw', window=0, **kwargs) -> pd.Series:
fw = False
bw = False
f = flagger.isFlagged(flags) & mask
if isinstance(window, int):
x = f.rolling(window=window + 1).sum()
if direction in ['fw', 'both']:
fw = x.fillna(method='bfill').astype(bool)
if direction in ['bw', 'both']:
bw = x.shift(-window).fillna(method='bfill').astype(bool)
else:
# time-based windows
if direction in ['bw', 'both']:
raise NotImplementedError
fw = f.rolling(window=window, closed='both').sum().astype(bool)
fmask = bw | fw
flags[fmask] = flagger.setFlag(flags[fmask], **kwargs)
return flags
def flagPeriod(flagger, flags, mask=True, flag_period=0, **kwargs) -> pd.Series:
return flagWindow(flagger, flags, mask, 'fw', window=flag_period, **kwargs)
def flagNext(flagger, flags, mask=True, flag_values=0, **kwargs) -> pd.Series:
return flagWindow(flagger, flags, mask, 'fw', window=flag_values, **kwargs)
def runner(meta, flagger, data, flags=None, nodata=np.nan):
if flags is None:
flags = pd.DataFrame(index=data.index)
# the required meta data columns
fields = [Fields.VARNAME, Fields.STARTDATE, Fields.ENDDATE, Fields.ASSIGN]
# NOTE:
# get to know every variable from meta
# should go into a separate function
for idx, configrow in meta.iterrows():
varname, _, _, assign = configrow[fields]
if varname not in flags and \
(varname in data or varname not in data and assign is True):
col_flags = flagger.initFlags(pd.DataFrame(index=data.index,
columns=[varname]))
flags = col_flags if flags.empty else flags.join(col_flags)
# NOTE:
# the outer loop runs over the flag tests, the inner one over the
# variables. Switching the loop order would complicate the
# reference to flags from other variables within the dataset
flag_fields = meta.columns.to_series().filter(regex=Fields.FLAGS)
for flag_pos, flag_field in enumerate(flag_fields):
# NOTE: just an optimization
if meta[flag_field].dropna().empty:
continue
for idx, configrow in meta.iterrows():
flag_test = configrow[flag_field]
if pd.isnull(flag_test):
continue
varname, start_date, end_date, _ = configrow[fields]
func_name, flag_params = parseFlag(flag_test)
if varname not in data and varname not in flags:
continue
dchunk = data.loc[start_date:end_date].copy()
if dchunk.empty:
continue
fchunk = flags.loc[start_date:end_date].copy()
try:
dchunk, fchunk = flagDispatch(func_name,
dchunk, fchunk, varname,
flagger, nodata=nodata,
**flag_params)
except NameError:
raise NameError(
f"function name {func_name} is not definied (variable '{varname}, 'line: {idx + 1})")
old = flagger.getFlags(flags.loc[start_date:end_date, varname])
new = flagger.getFlags(fchunk[varname])
mask = old != new
# flag a timespan after the condition is met
if Params.FLAGPERIOD in flag_params:
fchunk[varname] = flagPeriod(flagger, fchunk[varname], mask, **flag_params)
# flag a certain amount of values after condition is met
if Params.FLAGVALUES in flag_params:
fchunk[varname] = flagNext(flagger, fchunk[varname], mask, **flag_params)
if Params.FLAGPERIOD in flag_params or Params.FLAGVALUES in flag_params:
# hack as assignment above don't preserve categorical type
fchunk = fchunk.astype({
c: flagger.flags for c in fchunk.columns if flagger.flag_fields[0] in c})
if Params.PLOT in flag_params:
new = flagger.getFlags(fchunk[varname])
mask = old != new
plot(dchunk, fchunk, mask, varname, flagger, title=flag_test)
data.loc[start_date:end_date] = dchunk
flags[start_date:end_date] = fchunk.squeeze()
flagger.nextTest()
return data, flags
def plot(data, flags, flagmask, varname, flagger, interactive_backend=True, title="Data Plot"):
# the flagmask is True for flags to be shown False otherwise
if not interactive_backend:
# Import plot libs without interactivity, if not needed. This ensures that this can
# produce an plot.png even if tkinter is not installed. E.g. if one want to run this
# on machines without X-Server aka. graphic interface.
mpl.use('Agg')
else:
mpl.use('TkAgg')
from matplotlib import pyplot as plt
# needed for datetime conversion
from pandas.plotting import register_matplotlib_converters
register_matplotlib_converters()
def plot_vline(plt, points, color='blue'):
# workaround for ax.vlines() as this work unexpected
for point in points:
plt.axvline(point, color=color, linestyle=':')
def _plot(varname, ax):
x = data.index
y = data[varname]
flags_ = flags[varname]
nrofflags = len(flagger.flags.categories)
ax.plot(x, y, '-',markersize=1, color='silver')
if nrofflags == 3:
colors = {0:'silver', 1:'lime', 2:'red'}
if nrofflags == 4:
colors = {0:'silver', 1:'lime', 2:'yellow', 3:'red'}
# plot (all) data in silver
ax.plot(x, y, '-', color='silver', label='data')
# plot (all) missing data in silver
nans = y.isna()
ylim = plt.ylim()
flagged = flagger.isFlagged(flags_)
idx = y.index[nans & ~flagged]
# ax.vlines(idx, *ylim, linestyles=':', color='silver', label="missing")
plot_vline(ax, idx, color='silver')
# plot all flagged data in black
ax.plot(x[flagged], y[flagged], '.', color='black', label="flagged by other test")
# plot all flagged missing data (flagged before) in black
idx = y.index[nans & flagged & ~flagmask]
# ax.vlines(idx, *ylim, linestyles=':', color='black')
plot_vline(ax, idx, color='black')
ax.set_ylabel(varname)
# plot currently flagged data in color of flag
for i, f in enumerate(flagger.flags):
if i == 0:
continue
flagged = flagger.isFlagged(flags_, flag=f) & flagmask
label = f"flag: {f}" if i else 'data'
ax.plot(x[flagged], y[flagged], '.', color=colors[i], label=label)
idx = y.index[nans & flagged]
# ax.vlines(idx, *ylim, linestyles=':', color=colors[i])
plot_vline(ax, idx, color=colors[i])
if not isinstance(varname, (list, set)):
varname = set([varname])
plots = len(varname)
if plots > 1:
fig, axes = plt.subplots(plots, 1, sharex=True)
axes[0].set_title(title)
for i, v in enumerate(varname):
_plot(v, axes[i])
else:
fig, ax = plt.subplots()
plt.title(title)
_plot(varname.pop(), ax)
plt.xlabel('time')
# dummy plot for label `missing` see plot_vline for more info
plt.plot([], [], ':', color='silver', label="missing data")
plt.legend()
plt.show()
def prepareMeta(meta, data):
# NOTE: an option needed to only pass tests within an file and deduce
# everything else from data
# no dates given, fall back to the available date range
if Fields.STARTDATE not in meta:
meta = meta.assign(**{Fields.STARTDATE: np.nan})
if Fields.ENDDATE not in meta:
meta = meta.assign(**{Fields.ENDDATE: np.nan})
meta = meta.fillna(
{Fields.ENDDATE: data.index.max(),
Fields.STARTDATE: data.index.min()})
if Fields.ASSIGN not in meta:
meta = meta.assign(**{Fields.ASSIGN: False})
# rows without a variables name don't help much
meta = meta.dropna(subset=[Fields.VARNAME])
meta[Fields.STARTDATE] = pd.to_datetime(meta[Fields.STARTDATE])
meta[Fields.ENDDATE] = pd.to_datetime(meta[Fields.ENDDATE])
return meta
def readData(fname, index_col, nans):
data = pd.read_csv(
fname, index_col=index_col, parse_dates=True,
na_values=nans, low_memory=False)
data.columns = [c.split(" ")[0] for c in data.columns]
data = data.reindex(
pd.date_range(data.index.min(), data.index.max(), freq="10min"))
return data
if __name__ == "__main__":
datafname = "resources/data.csv"
metafname = "resources/meta.csv"
data = readData(datafname,
index_col="Date Time",
nans=["-9999", "-9999.0"])
meta = prepareMeta(pd.read_csv(metafname), data)
flagger = PositionalFlagger()
pdata, pflags = runner(meta, flagger, data)