Skip to content

Commit

Permalink
Release Process : Profiling and Warning Hunts #195. Avoid row-based a…
Browse files Browse the repository at this point in the history 
…pply for signal transformations. Some syntactic changes (sig instead of df)
  • Loading branch information
@antoinecarme
antoinecarme committed Apr 9, 2022
1 parent b44f18c commit 1ab3bba
Showing 1 changed file with 72 additions and 68 deletions.
140 changes: 72 additions & 68 deletions pyaf/TS/Signal_Transformation.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -68,14 +68,16 @@ def fit_scaling_params(self, sig):
pass

def scale_value(self, x):
if(np.fabs(self.mInputValueRange) < 1e-10):
return 0.0
return (x - self.mMinInputValue) / self.mInputValueRange;

def scale_signal(self, sig):
if(self.mScaling is not None):
# print("SCALE_START", sig.values[1:5]);
sig1 = sig.apply(self.scale_value);
sig1 = None
if(np.fabs(self.mInputValueRange) < 1e-10):
sig1 = 0.0
else:
sig1 = self.scale_value(sig)
# print("SCALE_END", sig1.values[1:5]);
return sig1;
else:
Expand All @@ -91,7 +93,7 @@ def rescale_value(self, x):
def rescale_signal(self, sig1):
if(self.mScaling is not None):
# print("RESCALE_START", sig1.values[1:5]);
sig = sig1.apply(self.rescale_value);
sig = self.rescale_value(sig1);
# print("RESCALE_END", sig.values[1:5]);
return sig;
else:
Expand Down Expand Up @@ -134,12 +136,12 @@ def test(self):
# testTransform(tr1);
pass

def dump_apply_invert(self, df_before_apply, df_after_apply):
df = pd.DataFrame(index = None);
df['before_apply'] = df_before_apply;
df['after_apply'] = df_after_apply;
print("dump_apply_invert_head", df.head());
print("dump_apply_invert_tail", df.tail());
def dump_apply_invert(self, sig_before_apply, sig_after_apply):
sig = pd.Series(index = None);
sig['before_apply'] = sig_before_apply;
sig['after_apply'] = sig_after_apply;
print("dump_apply_invert_head", sig.head());
print("dump_apply_invert_tail", sig.tail());

def check_not_nan(self, sig , name):
if(np.isnan(sig).any()):
Expand All @@ -163,11 +165,11 @@ def get_name(self, iSig):
def specific_fit(self , sig):
pass

def specific_apply(self, df):
return df;
def specific_apply(self, sig):
return sig;

def specific_invert(self, df):
return df;
def specific_invert(self, sig):
return sig;

def dump_values(self):
logger = tsutil.get_pyaf_logger();
Expand All @@ -192,10 +194,10 @@ def specific_fit(self , sig):
def specific_apply(self, sig):
return sig.cumsum(axis = 0)

def specific_invert(self, df):
df_orig = df - df.shift(1);
df_orig.iloc[0] = df.iloc[0];
return df_orig;
def specific_invert(self, sig):
sig_orig = sig - sig.shift(1);
sig_orig.iloc[0] = sig.iloc[0];
return sig_orig;

def dump_values(self):
logger = tsutil.get_pyaf_logger();
Expand Down Expand Up @@ -231,8 +233,8 @@ def signal2quant(self, x):
curve = self.mCurve;
return min(curve.keys(), key=lambda y:abs(float(curve[y])-x))

def specific_apply(self, df):
lSignal_Q = df.apply(self.signal2quant);
def specific_apply(self, sig):
lSignal_Q = sig.apply(self.signal2quant);
return lSignal_Q;

def quant2signal(self, x):
Expand All @@ -245,8 +247,8 @@ def quant2signal(self, x):
val = curve[key]
return val;

def specific_invert(self, df):
lSignal = df.apply(self.quant2signal);
def specific_invert(self, sig):
lSignal = sig.apply(self.quant2signal);
return lSignal;

def dump_values(self):
Expand All @@ -273,29 +275,30 @@ def specific_fit(self, sig):
pass


def specific_apply(self, df):
def specific_apply(self, sig):
lEps = 1e-3
assert(df.min() > -lEps)
log_df = df.apply(lambda x : np.log(max(x , lEps)));
assert(sig.min() > -lEps)
log_sig = np.log(sig.clip(lEps, None))
if(abs(self.mLambda) <= 0.001):
return log_df;
return log_sig;
lLimit = 5.0 / abs(self.mLambda)
log_df = log_df.clip(-lLimit , lLimit)
df1 = (np.exp(log_df * self.mLambda) - 1) / self.mLambda
return df1;
log_sig = log_sig.clip(-lLimit , lLimit)
sig1 = (np.exp(log_sig * self.mLambda) - 1) / self.mLambda
return sig1;

def invert_value(self, y):
x = y;
lEps = 1e-5
x1 = np.log(max(self.mLambda * x + 1, lEps)) / self.mLambda;
x0 = self.mLambda * x + 1
x1 = np.log(x0.clip(lEps, None)) / self.mLambda;
return np.exp(x1).clip(0, 1) ;

def specific_invert(self, df):
def specific_invert(self, sig):
if(abs(self.mLambda) <= 0.001):
df_orig = np.exp(df).clip(0, 1);
return df_orig;
df_pos = df.apply(self.invert_value);
return df_pos;
sig_orig = np.exp(sig).clip(0, 1);
return sig_orig;
sig_pos = self.invert_value(sig)
return sig_pos;

def dump_values(self):
logger = tsutil.get_pyaf_logger();
Expand All @@ -321,16 +324,16 @@ def specific_fit(self, sig):
pass


def specific_apply(self, df):
df_shifted = df.shift(1)
df_shifted.iloc[0] = self.mFirstValue;
lResult = df - df_shifted
def specific_apply(self, sig):
sig_shifted = sig.shift(1)
sig_shifted.iloc[0] = self.mFirstValue;
lResult = sig - sig_shifted
return lResult

def specific_invert(self, df):
df_cumsum = df.cumsum();
df_orig = df_cumsum + self.mFirstValue;
return df_orig;
def specific_invert(self, sig):
sig_cumsum = sig.cumsum();
sig_orig = sig_cumsum + self.mFirstValue;
return sig_orig;

def dump_values(self):
logger = tsutil.get_pyaf_logger();
Expand All @@ -354,16 +357,17 @@ def specific_fit(self, sig):
self.mFirstValue = sig.iloc[0];
pass

def specific_apply(self, df):
def specific_apply(self, sig):
lEps = 1e-2
# print("RelDiff_apply_DEBUG_START" , self.mFirstValue, df.values[0:10]);
df1 = df.apply(lambda x : x if (abs(x) > lEps) else lEps)
df_shifted = df1.shift(1)
# df_shifted[df_shifted <= lEps] = lEps
rate = (df1 - df_shifted) / df_shifted
# print("RelDiff_apply_DEBUG_START" , self.mFirstValue, sig.values[0:10]);
sig1 = sig
sig1[sig1.abs() <= lEps] = lEps
sig_shifted = sig1.shift(1)
# sig_shifted[sig_shifted <= lEps] = lEps
rate = (sig1 - sig_shifted) / sig_shifted
rate.iloc[0] = 0.0;
# print(df1)
# print(df_shifted)
# print(sig1)
# print(sig_shifted)
rate = rate.clip(-1.0e+2 , +1.0e+2)
# print("RelDiff_apply_DEBUG_END" , rate[0:10]);
return rate;
Expand All @@ -377,17 +381,17 @@ def cumprod_no_overflow(self, rate):
lResult = np.exp(lCumSum)
return lResult

def specific_invert(self, df):
# print("RelDiff_invert_DEBUG_START" , self.mFirstValue, df.values[0:10]);
rate = df + 1;
def specific_invert(self, sig):
# print("RelDiff_invert_DEBUG_START" , self.mFirstValue, sig.values[0:10]);
rate = sig + 1;
rate = rate.clip(-1.0e+8 , +1.0e+8)
rate_cum = self.cumprod_no_overflow(rate);
df_orig = rate_cum.clip(-1.0e+8 , +1.0e+8)
df_orig = self.mFirstValue * df_orig;
sig_orig = rate_cum.clip(-1.0e+8 , +1.0e+8)
sig_orig = self.mFirstValue * sig_orig;
# print("rate" , rate)
# print("rate_cum", rate_cum)
# print("RelDiff_invert_DEBUG_START" , df_orig[0:10])
return df_orig;
# print("RelDiff_invert_DEBUG_START" , sig_orig[0:10])
return sig_orig;

def dump_values(self):
logger = tsutil.get_pyaf_logger();
Expand Down Expand Up @@ -424,15 +428,15 @@ def inv_logit(self, y):
p = x / (1 + x);
return p;

def specific_apply(self, df):
def specific_apply(self, sig):
# logit
df1 = df.apply(self.logit);
return df1;
sig1 = self.logit(sig)
return sig1;

def specific_invert(self, df):
def specific_invert(self, sig):
# logit
df1 = df.apply(self.inv_logit);
return df1;
sig1 = self.inv_logit(sig)
return sig1;

def dump_values(self):
logger = tsutil.get_pyaf_logger();
Expand Down Expand Up @@ -461,12 +465,12 @@ def specific_fit(self , sig):
pass

def specific_apply(self, sig):
y = sig.apply(lambda x : 2 * np.sqrt(x + self.mConstant));
y = 2 * np.sqrt(sig + self.mConstant)
return y;

def specific_invert(self, sig):
y1 = sig.clip(1.22, 2.34)
x = y1.apply(lambda x : ((x/2 * x/2) - self.mConstant))
x = (y1/2 * y1/2) - self.mConstant
return x;

def dump_values(self):
Expand Down Expand Up @@ -494,11 +498,11 @@ def specific_fit(self , sig):

def specific_apply(self, sig):
eps = 1.0e-2;
y = sig.apply(lambda x : np.arctanh(np.clip(x , -1 + eps , 1.0 - eps)));
y = np.arctanh(np.clip(sig.values , -1 + eps , 1.0 - eps))
return y;

def specific_invert(self, sig):
x = sig.apply(np.tanh);
x = np.tanh(sig.values);
return x;

def dump_values(self):
Expand Down

0 comments on commit 1ab3bba

Please sign in to comment.