ovos-backend-manager/ovos_backend_manager/metrics.py
2023-04-08 00:50:24 +01:00

602 lines
21 KiB
Python

import json
import os
import time
from cutecharts.charts import Pie, Bar, Scatter
from ovos_backend_manager.configuration import CONFIGURATION, DB
from pywebio.input import actions
from pywebio.output import put_text, popup, put_code, put_markdown, put_html, use_scope, put_image
chart_type = Pie
def device_select(back_handler=None):
devices = {device["uuid"]: f"{device['name']}@{device['device_location']}"
for device in DB.list_devices()}
buttons = [{'label': "All Devices", 'value': "all"}] + \
[{'label': d, 'value': uuid} for uuid, d in devices.items()]
if back_handler:
buttons.insert(0, {'label': '<- Go Back', 'value': "main"})
if devices:
uuid = actions(label="What device would you like to inspect?",
buttons=buttons)
if uuid == "main":
metrics_menu(back_handler=back_handler)
return
else:
if uuid == "all":
uuid = None
if uuid is not None:
with use_scope("main_view", clear=True):
put_markdown(f"\nDevice: {uuid}")
metrics_menu(uuid=uuid, back_handler=back_handler)
else:
popup("No devices paired yet!")
metrics_menu(back_handler=back_handler)
def metrics_select(back_handler=None, uuid=None):
buttons = []
metrics = DB.list_metrics()
if not len(metrics):
with use_scope("main_view", clear=True):
put_text("No metrics uploaded yet!")
metrics_menu(back_handler=back_handler, uuid=uuid)
return
for m in metrics:
name = f"{m['metric_id']}-{m['metric_type']}"
if uuid is not None and m["uuid"] != uuid:
continue
buttons.append({'label': name, 'value': m['metric_id']})
if back_handler:
buttons.insert(0, {'label': '<- Go Back', 'value': "main"})
metric_id = actions(label="Select a metric to inspect",
buttons=buttons)
if metric_id == "main":
device_select(back_handler=back_handler)
return
with use_scope("main_view", clear=True):
put_markdown("# Metadata")
put_code(json.dumps(metric_id, indent=4), "json")
metrics_select(back_handler=back_handler, uuid=uuid)
def _plot_metrics(uuid, selected_metric="types"):
if uuid is not None:
m = DeviceMetricsReportGenerator(uuid)
else:
m = MetricsReportGenerator()
with use_scope("main_view", clear=True):
if uuid is not None:
put_markdown(f"\nDevice: {uuid}")
if selected_metric == "timings":
put_html(m.timings_chart().render_notebook())
elif selected_metric == "stt":
silents = max(0, m.total_stt - m.total_utt)
put_markdown(f"""Total Transcriptions: {m.total_stt}
Total Recording uploads: {m.total_utt}
Silent Activations (estimate): {silents}""")
if chart_type == Pie:
put_html(m.stt_pie_chart().render_notebook())
else:
put_html(m.stt_bar_chart().render_notebook())
elif selected_metric == "devices":
md = f"""# Devices Report
Total Devices: {m.total_devices}
Total untracked: {len(m.untracked_devices)}
Total active (estimate): {len(m.active_devices)}
Total dormant (estimate): {len(m.dormant_devices)}"""
put_markdown(md)
if chart_type == Pie:
put_html(m.devices_pie_chart().render_notebook())
else:
put_html(m.devices_bar_chart().render_notebook())
elif selected_metric == "intents":
txt_estimate = max(m.total_intents + m.total_fallbacks - m.total_stt, 0)
stt_estimate = max(m.total_intents + m.total_fallbacks - txt_estimate, 0)
md = f"""# Intent Matches Report
Total queries: {m.total_intents + m.total_fallbacks}
Total text queries (estimate): {txt_estimate}
Total speech queries (estimate): {stt_estimate}
Total Matches: {m.total_intents}"""
put_markdown(md)
if chart_type == Bar:
put_html(m.intents_bar_chart().render_notebook())
else:
put_html(m.intents_pie_chart().render_notebook())
elif selected_metric == "ww":
bad = max(0, m.total_stt - m.total_ww)
silents = max(0, m.total_stt - m.total_utt)
put_markdown(f"""Total WakeWord uploads: {m.total_ww}
Total WakeWord detections (estimate): {m.total_stt}
False Activations (estimate): {bad or silents}
Silent Activations (estimate): {silents}""")
if chart_type == Pie:
put_html(m.ww_pie_chart().render_notebook())
else:
put_html(m.ww_bar_chart().render_notebook())
elif selected_metric == "tts":
if chart_type == Pie:
put_html(m.tts_pie_chart().render_notebook())
else:
put_html(m.tts_bar_chart().render_notebook())
elif selected_metric == "types":
put_markdown(f"""
# Metrics Report
Total Intents: {m.total_intents}
Total Fallbacks: {m.total_fallbacks}
Total Transcriptions: {m.total_stt}
Total TTS: {m.total_tts}
""")
if chart_type == Pie:
put_html(m.metrics_type_pie_chart().render_notebook())
else:
put_html(m.metrics_type_bar_chart().render_notebook())
elif selected_metric == "fallback":
f = 0
if m.total_intents + m.total_fallbacks > 0:
f = m.total_intents / (m.total_intents + m.total_fallbacks)
put_markdown(f"""
# Fallback Matches Report
Total queries: {m.total_intents + m.total_fallbacks}
Total Intents: {m.total_intents}
Total Fallbacks: {m.total_fallbacks}
Failure Percentage (estimate): {1 - f}
""")
if chart_type == Pie:
put_html(m.fallback_pie_chart().render_notebook())
else:
put_html(m.fallback_bar_chart().render_notebook())
elif selected_metric == "opt-in":
md = ""
if uuid is None:
md = f"""# Open Dataset Report
Total Registered Devices: {len(DB.list_devices())}
Currently Opted-in: {len([d for d in DB.list_devices() if d["opt_in"]])}
Unique Devices seen: {m.total_devices}"""
# Open Dataset Report"""
md += f"""
Total Metrics submitted: {m.total_metrics}
Total WakeWords submitted: {m.total_ww}
Total Utterances submitted: {m.total_utt}"""
put_markdown(md)
if chart_type == Pie:
put_html(m.dataset_pie_chart().render_notebook())
else:
put_html(m.dataset_bar_chart().render_notebook())
def metrics_menu(back_handler=None, uuid=None, selected_metric="types"):
global chart_type
with use_scope("logo", clear=True):
img = open(f'{os.path.dirname(__file__)}/res/metrics.png', 'rb').read()
put_image(img)
_plot_metrics(uuid, selected_metric)
buttons = [{'label': 'Timings', 'value': "timings"},
{'label': 'Metric Types', 'value': "types"},
{'label': 'Intents', 'value': "intents"},
{'label': 'FallbackSkill', 'value': "fallback"},
{'label': 'STT', 'value': "stt"},
{'label': 'TTS', 'value': "tts"},
{'label': 'Wake Words', 'value': "ww"},
{'label': 'Open Dataset', 'value': "opt-in"}]
if chart_type == Pie:
buttons.append({'label': 'Bar style graphs', 'value': "chart"})
elif chart_type == Bar:
buttons.append({'label': 'Pie style graphs', 'value': "chart"})
if uuid is not None:
buttons.append({'label': 'Delete Device metrics', 'value': "delete_metrics"})
else:
buttons.insert(1, {'label': 'Devices', 'value': "devices"})
buttons.append({'label': 'Inspect Devices', 'value': "metrics"})
buttons.append({'label': 'Delete ALL metrics', 'value': "delete_metrics"})
if back_handler:
buttons.insert(0, {'label': '<- Go Back', 'value': "main"})
opt = actions(label="What would you like to do?",
buttons=buttons)
if opt == "chart":
if chart_type == Pie:
chart_type = Bar
else:
chart_type = Pie
elif opt in ["devices", "intents", "stt", "ww", "tts", "types", "fallback", "opt-in", "timings"]:
selected_metric = opt
elif opt == "metrics":
device_select(back_handler=back_handler)
elif opt == "delete_metrics":
if uuid is not None:
with use_scope("main_view", clear=True):
put_markdown(f"\nDevice: {uuid}")
with popup("Are you sure you want to delete the metrics database?"):
put_text("this can not be undone, proceed with caution!")
put_text("ALL metrics will be lost")
opt = actions(label="Delete metrics database?",
buttons=[{'label': "yes", 'value': True},
{'label': "no", 'value': False}])
if opt:
for m in DB.list_metrics():
DB.delete_metric(m["metric_id"])
with use_scope("main_view", clear=True):
if back_handler:
back_handler()
else:
metrics_menu(back_handler=back_handler, uuid=uuid,
selected_metric=selected_metric)
return
elif opt == "main":
with use_scope("main_view", clear=True):
if uuid is not None:
device_select(back_handler=back_handler)
elif back_handler:
back_handler()
return
metrics_menu(back_handler=back_handler, uuid=uuid,
selected_metric=selected_metric)
class MetricsReportGenerator:
def __init__(self):
self.total_intents = 0
self.total_fallbacks = 0
self.total_stt = 0
self.total_tts = 0
self.total_ww = len(DB.list_ww_recordings())
self.total_utt = len(DB.list_stt_recordings())
self.total_devices = len(DB.list_devices())
self.total_metrics = len(DB.list_metrics())
self.intents = {}
self.fallbacks = {}
self.ww = {}
self.tts = {}
self.stt = {}
self.devices = {}
self.stt_timings = []
self.tts_timings = []
self.intent_timings = []
self.fallback_timings = []
self.device_timings = []
self.load_metrics()
def reset_metrics(self):
self.total_intents = 0
self.total_fallbacks = 0
self.total_stt = 0
self.total_tts = 0
self.total_ww = len(DB.list_ww_recordings())
self.total_utt = len(DB.list_stt_recordings())
self.total_devices = 0
self.total_metrics = len(DB.list_metrics())
self.intents = {}
self.devices = {}
self.fallbacks = {}
self.tts = {}
self.stt = {}
self.ww = {}
self.stt_timings = []
self.tts_timings = []
self.intent_timings = []
self.fallback_timings = []
self.device_timings = []
def load_metrics(self):
self.reset_metrics()
for m in DB.list_metrics():
if m["uuid"] not in self.devices:
self.total_devices += 1
self._process_metric(m)
for ww in DB.list_ww_recordings():
if ww["meta"]["name"] not in self.ww:
self.ww[ww["meta"]["name"]] = 0
else:
self.ww[ww["meta"]["name"]] += 1
@property
def active_devices(self):
thresh = time.time() - 7 * 24 * 60 * 60
return [uuid for uuid, ts in self.devices.items()
if ts > thresh and uuid not in self.untracked_devices]
@property
def dormant_devices(self):
return [uuid for uuid in self.devices.keys()
if uuid not in self.untracked_devices
and uuid not in self.active_devices]
@property
def untracked_devices(self):
return [dev["uuid"] for dev in DB.list_devices() if not dev["opt_in"]]
# cute charts
def timings_chart(self):
chart = Scatter("Execution Time")
chart.set_options(y_tick_count=8, is_show_line=True,
x_label="Unix Time", y_label="Seconds")
chart.add_series(
"STT", [(z[0], z[1]) for z in self.stt_timings]
)
chart.add_series(
"TTS", [(z[0], z[1]) for z in self.tts_timings]
)
chart.add_series(
"Intent Matching", [(z[0], z[1]) for z in self.intent_timings]
)
chart.add_series(
"Fallback Handling", [(z[0], z[1]) for z in self.fallback_timings]
)
return chart
def devices_pie_chart(self):
chart = Pie("Devices")
chart.set_options(
labels=["active", "dormant", "untracked"],
inner_radius=0,
)
chart.add_series([len(self.active_devices),
len(self.dormant_devices),
len(self.untracked_devices)])
return chart
def devices_bar_chart(self):
chart = Bar("Devices")
chart.set_options(
labels=["active", "dormant", "untracked"],
x_label="Status", y_label="Number"
)
chart.add_series("Count", [len(self.active_devices),
len(self.dormant_devices),
len(self.untracked_devices)])
return chart
def ww_bar_chart(self):
chart = Bar("Wake Words")
labels = []
series = []
for ww, count in self.ww.items():
labels.append(ww)
series.append(count)
chart.set_options(
labels=labels, x_label="Wake Word", y_label="# Submitted"
)
chart.add_series("Count", series)
return chart
def ww_pie_chart(self):
chart = Pie("Wake Words")
labels = []
series = []
for ww, count in self.ww.items():
labels.append(ww)
series.append(count)
chart.set_options(
labels=labels,
inner_radius=0,
)
chart.add_series(series)
return chart
def dataset_pie_chart(self):
chart = Pie("Uploaded Data")
chart.set_options(
labels=["wake-words", "utterances", "metrics"],
inner_radius=0,
)
chart.add_series([self.total_ww, self.total_utt, self.total_metrics])
return chart
def dataset_bar_chart(self):
chart = Bar("Uploaded Data")
chart.set_options(
labels=["wake-words", "utterances", "metrics"],
x_label="Data Type", y_label="# Submitted"
)
chart.add_series("Count", [self.total_ww, self.total_utt, self.total_metrics])
return chart
def metrics_type_bar_chart(self):
chart = Bar("Metric Types")
chart.set_options(
labels=["intents", "fallbacks", "stt", "tts"],
x_label="Metric Type", y_label="# Submitted"
)
chart.add_series("Number", [self.total_intents,
self.total_fallbacks,
self.total_stt,
self.total_tts])
return chart
def metrics_type_pie_chart(self):
chart = Pie("Metric Types")
chart.set_options(
labels=["intents", "fallbacks", "stt", "tts"],
inner_radius=0,
)
chart.add_series([self.total_intents,
self.total_fallbacks,
self.total_stt,
self.total_tts])
return chart
def intents_bar_chart(self):
chart = Bar("Intent Matches")
chart.set_options(labels=list(self.intents.keys()),
x_label="Intent Name", y_label="Times Triggered")
chart.add_series("Count", list(self.intents.values()))
return chart
def intents_pie_chart(self):
chart = Pie("Intent Matches")
chart.set_options(
labels=list(self.intents.keys()),
inner_radius=0,
)
chart.add_series(list(self.intents.values()))
return chart
def fallback_bar_chart(self):
chart = Bar("Fallback Skills")
chart.set_options(
labels=list(self.fallbacks.keys()),
x_label="Fallback Handler", y_label="Times Triggered"
)
chart.add_series("Count", list(self.fallbacks.values()))
return chart
def fallback_pie_chart(self):
chart = Pie("Fallback Skills")
chart.set_options(
labels=list(self.fallbacks.keys()),
inner_radius=0,
)
chart.add_series(list(self.fallbacks.values()))
return chart
def tts_bar_chart(self):
chart = Bar("Text To Speech Engines")
chart.set_options(
labels=list(self.tts.keys()),
x_label="Engine", y_label="Times Triggered"
)
chart.add_series("Count", list(self.tts.values()))
return chart
def tts_pie_chart(self):
chart = Pie("Text To Speech Engines")
chart.set_options(
labels=list(self.tts.keys()),
inner_radius=0,
)
chart.add_series(list(self.tts.values()))
return chart
def stt_bar_chart(self):
chart = Bar("Speech To Text Engines")
chart.set_options(
labels=list(self.stt.keys()),
x_label="Engine", y_label="Times Triggered"
)
chart.add_series("Count", list(self.stt.values()))
return chart
def stt_pie_chart(self):
chart = Pie("Speech To Text Engines")
chart.set_options(
labels=list(self.stt.keys()),
inner_radius=0,
)
chart.add_series(list(self.stt.values()))
return chart
def _process_metric(self, m):
start = m["meta"]["start_time"]
end = m["meta"]["time"]
duration = end - start
if m["uuid"] not in self.devices or \
m["meta"]["time"] > self.devices[m["uuid"]]:
self.devices[m["uuid"]] = m["meta"]["time"]
if m["metric_type"] == "intent_service":
label = m["meta"]["intent_type"]
self.intent_timings.append((start, duration, label))
self.total_intents += 1
k = f"{m['meta']['intent_type']}"
if k not in self.intents:
self.intents[k] = 0
self.intents[k] += 1
if m["metric_type"] == "fallback_handler":
self.total_fallbacks += 1
k = f"{m['meta']['handler']}"
if m['meta'].get("skill_id"):
k = f"{m['meta']['skill_id']}:{m['meta']['handler']}"
if k not in self.fallbacks:
self.fallbacks[k] = 0
self.fallbacks[k] += 1
label = k
self.fallback_timings.append((start, duration, label))
if m["metric_type"] == "stt":
label = m["meta"]["transcription"]
self.stt_timings.append((start, duration, label))
self.total_stt += 1
k = f"{m['meta']['stt']}"
if k not in self.stt:
self.stt[k] = 0
self.stt[k] += 1
if m["metric_type"] == "speech":
label = m["meta"]["utterance"]
self.tts_timings.append((start, duration, label))
self.total_tts += 1
k = f"{m['meta']['tts']}"
if k not in self.tts:
self.tts[k] = 0
self.tts[k] += 1
self.device_timings.append((start, duration, m["uuid"]))
# sort by timestamp
self.device_timings = sorted(self.device_timings, key=lambda k: k[0], reverse=True)
self.stt_timings = sorted(self.stt_timings, key=lambda k: k[0], reverse=True)
self.tts_timings = sorted(self.tts_timings, key=lambda k: k[0], reverse=True)
self.intent_timings = sorted(self.intent_timings, key=lambda k: k[0], reverse=True)
self.fallback_timings = sorted(self.fallback_timings, key=lambda k: k[0], reverse=True)
class DeviceMetricsReportGenerator(MetricsReportGenerator):
def __init__(self, uuid):
self.uuid = uuid
super().__init__()
def load_metrics(self):
self.reset_metrics()
self.total_ww = len([ww for ww in DB.list_ww_recordings()
if ww["uuid"] == self.uuid])
self.total_metrics = 0
self.total_utt = len([utt for utt in DB.list_stt_recordings()
if utt["uuid"] == self.uuid])
for m in DB.list_metrics():
if m["uuid"] != self.uuid:
continue
self._process_metric(m)
self.total_metrics += 1
for ww in DB.list_ww_recordings():
if ww["uuid"] != self.uuid:
continue
if ww["meta"]["name"] not in self.ww:
self.ww[ww["meta"]["name"]] = 0
else:
self.ww[ww["meta"]["name"]] += 1
if __name__ == "__main__":
for ww in DB.list_ww_recordings():
print(ww)