In [ ]:
import sys,os
sys.path.append("..")
import numpy.random
import pandas as pd
import shutil
import tempfile
import trappy
trace_thermal = "./trace.txt"
trace_sched = "../tests/raw_trace.dat"
TEMP_BASE = "/tmp"
In [ ]:
def setup_thermal():
tDir = tempfile.mkdtemp(dir="/tmp", prefix="trappy_doc", suffix = ".tempDir")
shutil.copyfile(trace_thermal, os.path.join(tDir, "trace.txt"))
return tDir
def setup_sched():
tDir = tempfile.mkdtemp(dir="/tmp", prefix="trappy_doc", suffix = ".tempDir")
shutil.copyfile(trace_sched, os.path.join(tDir, "trace.dat"))
return tDir
In [ ]:
temp_thermal_location = setup_thermal()
trace1 = trappy.FTrace(temp_thermal_location)
trace2 = trappy.FTrace(temp_thermal_location)
trace2.thermal.data_frame["temp"] = trace1.thermal.data_frame["temp"] * 2
trace2.cpu_out_power.data_frame["power"] = trace1.cpu_out_power.data_frame["power"] * 2
Interactive Line Plots Supports the same API as the LinePlot but provide an interactive plot that can be zoomed by clicking and dragging on the desired area. Double clicking resets the zoom.
We can create an interactive plot easily from a dataframe by passing the data frame and the columns we want to plot as parameters:
In [ ]:
columns = ["tick", "tock"]
df = pd.DataFrame(numpy.random.randn(1000, 2), columns=columns).cumsum()
trappy.ILinePlot(df, column=columns).view()
It is also possible to plot traces with different index values (i.e. x-axis values)
In [ ]:
columns = ["tick", "tock", "bang"]
df_len = 1000
df1 = pd.DataFrame(numpy.random.randn(df_len, 3), columns=columns, index=range(df_len)).cumsum()
df2 = pd.DataFrame(numpy.random.randn(df_len, 3), columns=columns, index=(numpy.arange(0.5, df_len, 1))).cumsum()
In [ ]:
trappy.ILinePlot([df1, df2], column="tick").view()
This does not affect filtering or pivoting in any way
In [ ]:
df1["bang"] = df1["bang"].apply(lambda x: numpy.random.randint(0, 4))
df2["bang"] = df2["bang"].apply(lambda x: numpy.random.randint(0, 4))
In [ ]:
trappy.ILinePlot([df1, df2], column="tick", filters = {'bang' : [2]}, title="tick column values for which bang is 2").view()
In [ ]:
trappy.ILinePlot([df1, df2], column="tick", pivot="bang", title="tick column pivoted on bang column").view()
We can also create them from trace objects
In [ ]:
map_label = {
"00000000,00000006" : "A57",
"00000000,00000039" : "A53",
}
l = trappy.ILinePlot(
trace1, # TRAPpy FTrace Object
trappy.cpu_power.CpuInPower, # TRAPpy Event (maps to a unique word in the Trace)
column=[ # Column(s)
"dynamic_power",
"load1"],
filters={ # Filter the data
"cdev_state": [
1,
0]},
pivot="cpus", # One plot for each pivot will be created
map_label=map_label, # Optionally, provide an alternative label for pivots
per_line=1) # Number of graphs per line
l.view()
You can also change the drawstyle to "steps-post" for step plots. These are suited if the data is discrete and linear interploation is not required between two data points
In [ ]:
l = trappy.ILinePlot(
trace1, # TRAPpy FTrace Object
trappy.cpu_power.CpuInPower, # TRAPpy Event (maps to a unique word in the Trace)
column=[ # Column(s)
"dynamic_power",
"load1"],
filters={ # Filter the data
"cdev_state": [
1,
0]},
pivot="cpus", # One plot for each pivot will be created
per_line=1, # Number of graphs per line
drawstyle="steps-post")
l.view()
ILinePlots can zoom all at the same time. You can do so using the group and sync_zoom parameter. All ILinePlots using the same group name zoom at the same time.
In [ ]:
trappy.ILinePlot(
trace1,
signals=["cpu_in_power:dynamic_power", "cpu_in_power:load1"],
pivot="cpus",
group="synchronized",
sync_zoom=True
).view()
TRAPpy's Interactive Plotter features an Interactive Event TimeLine Plot. It accepts an input data of the type
{ "A" : [
[event_start, event_end, lane],
.
.
[event_start, event_end, lane],
],
.
.
.
"B" : [
[event_start, event_end, lane],
.
.
[event_start, event_end, lane],
.
.
.
}
Hovering on the rectangles gives the name of the process element and scrolling on the Plot Area and the window in the summary controls the zoom. One can also click and drag for panning a zoomed graph.
For Example:
In [ ]:
A = [
[0, 3, 0],
[4, 5, 2],
]
B = [
[0, 2, 1],
[2, 3, 3],
[3, 4, 0],
]
C = [
[0, 2, 3],
[2, 3, 2],
[3, 4, 1],
]
EVENTS = {}
EVENTS["A"] = A
EVENTS["B"] = B
EVENTS["C"] = C
trappy.EventPlot(EVENTS,
keys=EVENTS.keys, # Name of the Process Element
lane_prefix="LANE: ", # Name of Each TimeLine
num_lanes=4, # Number of Timelines
domain=[0,5] # Time Domain
).view()
Lane names can also be specified as strings (or hashable objects that have an str representation) as follows
In [ ]:
A = [
[0, 3, "zero"],
[4, 5, "two"],
]
B = [
[0, 2, 1],
[2, 3, "three"],
[3, 4, "zero"],
]
C = [
[0, 2, "three"],
[2, 3, "two"],
[3, 4, 1],
]
EVENTS = {}
EVENTS["A"] = A
EVENTS["B"] = B
EVENTS["C"] = C
trappy.EventPlot(EVENTS,
keys=EVENTS.keys, # Name of the Process Element
lanes=["zero", 1, "two", "three"],
domain=[0,5] # Time Domain
).view()
A specification of the EventPlot creates a kernelshark like plot if the sched_switch event is enabled in the traces
In [ ]:
f = setup_sched()
trappy.plotter.plot_trace(f)