aboutsummaryrefslogtreecommitdiffstats
path: root/resources/tools/dash/app/pal/utils/utils.py
diff options
context:
space:
mode:
authorpmikus <peter.mikus@protonmail.ch>2022-09-19 08:49:01 +0200
committerPeter Mikus <peter.mikus@protonmail.ch>2022-09-19 06:54:43 +0000
commitd6a60b5043c6f7c3dfc45853feb68d0aca5a4a5f (patch)
tree022343584d4e00a0da8ef12eb4e713d67344fc95 /resources/tools/dash/app/pal/utils/utils.py
parentd2ef7bc01df66f6a27f25d061db064cf4a463267 (diff)
feat(uti): Move directory
Signed-off-by: pmikus <peter.mikus@protonmail.ch> Change-Id: I7300ecfe756baaf3fbeedb020070f882cfaca445
Diffstat (limited to 'resources/tools/dash/app/pal/utils/utils.py')
-rw-r--r--resources/tools/dash/app/pal/utils/utils.py344
1 files changed, 0 insertions, 344 deletions
diff --git a/resources/tools/dash/app/pal/utils/utils.py b/resources/tools/dash/app/pal/utils/utils.py
deleted file mode 100644
index 9e4eeeb892..0000000000
--- a/resources/tools/dash/app/pal/utils/utils.py
+++ /dev/null
@@ -1,344 +0,0 @@
-# Copyright (c) 2022 Cisco and/or its affiliates.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at:
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-"""Function used by Dash applications.
-"""
-
-import pandas as pd
-import dash_bootstrap_components as dbc
-
-from numpy import isnan
-from dash import dcc
-from datetime import datetime
-
-from ..jumpavg import classify
-from ..utils.constants import Constants as C
-from ..utils.url_processing import url_encode
-
-
-def classify_anomalies(data):
- """Process the data and return anomalies and trending values.
-
- Gather data into groups with average as trend value.
- Decorate values within groups to be normal,
- the first value of changed average as a regression, or a progression.
-
- :param data: Full data set with unavailable samples replaced by nan.
- :type data: OrderedDict
- :returns: Classification and trend values
- :rtype: 3-tuple, list of strings, list of floats and list of floats
- """
- # NaN means something went wrong.
- # Use 0.0 to cause that being reported as a severe regression.
- bare_data = [0.0 if isnan(sample) else sample for sample in data.values()]
- # TODO: Make BitCountingGroupList a subclass of list again?
- group_list = classify(bare_data).group_list
- group_list.reverse() # Just to use .pop() for FIFO.
- classification = list()
- avgs = list()
- stdevs = list()
- active_group = None
- values_left = 0
- avg = 0.0
- stdv = 0.0
- for sample in data.values():
- if isnan(sample):
- classification.append("outlier")
- avgs.append(sample)
- stdevs.append(sample)
- continue
- if values_left < 1 or active_group is None:
- values_left = 0
- while values_left < 1: # Ignore empty groups (should not happen).
- active_group = group_list.pop()
- values_left = len(active_group.run_list)
- avg = active_group.stats.avg
- stdv = active_group.stats.stdev
- classification.append(active_group.comment)
- avgs.append(avg)
- stdevs.append(stdv)
- values_left -= 1
- continue
- classification.append("normal")
- avgs.append(avg)
- stdevs.append(stdv)
- values_left -= 1
- return classification, avgs, stdevs
-
-
-def get_color(idx: int) -> str:
- """Returns a color from the list defined in Constants.PLOT_COLORS defined by
- its index.
-
- :param idx: Index of the color.
- :type idx: int
- :returns: Color defined by hex code.
- :trype: str
- """
- return C.PLOT_COLORS[idx % len(C.PLOT_COLORS)]
-
-
-def show_tooltip(tooltips:dict, id: str, title: str,
- clipboard_id: str=None) -> list:
- """Generate list of elements to display a text (e.g. a title) with a
- tooltip and optionaly with Copy&Paste icon and the clipboard
- functionality enabled.
-
- :param tooltips: Dictionary with tooltips.
- :param id: Tooltip ID.
- :param title: A text for which the tooltip will be displayed.
- :param clipboard_id: If defined, a Copy&Paste icon is displayed and the
- clipboard functionality is enabled.
- :type tooltips: dict
- :type id: str
- :type title: str
- :type clipboard_id: str
- :returns: List of elements to display a text with a tooltip and
- optionaly with Copy&Paste icon.
- :rtype: list
- """
-
- return [
- dcc.Clipboard(target_id=clipboard_id, title="Copy URL") \
- if clipboard_id else str(),
- f"{title} ",
- dbc.Badge(
- id=id,
- children="?",
- pill=True,
- color="white",
- text_color="info",
- class_name="border ms-1",
- ),
- dbc.Tooltip(
- children=tooltips.get(id, str()),
- target=id,
- placement="auto"
- )
- ]
-
-
-def label(key: str) -> str:
- """Returns a label for input elements (dropdowns, ...).
-
- If the label is not defined, the function returns the provided key.
-
- :param key: The key to the label defined in Constants.LABELS.
- :type key: str
- :returns: Label.
- :rtype: str
- """
- return C.LABELS.get(key, key)
-
-
-def sync_checklists(options: list, sel: list, all: list, id: str) -> tuple:
- """Synchronize a checklist with defined "options" with its "All" checklist.
-
- :param options: List of options for the cheklist.
- :param sel: List of selected options.
- :param all: List of selected option from "All" checklist.
- :param id: ID of a checklist to be used for synchronization.
- :returns: Tuple of lists with otions for both checklists.
- :rtype: tuple of lists
- """
- opts = {v["value"] for v in options}
- if id =="all":
- sel = list(opts) if all else list()
- else:
- all = ["all", ] if set(sel) == opts else list()
- return sel, all
-
-
-def list_tests(selection: dict) -> list:
- """Transform list of tests to a list of dictionaries usable by checkboxes.
-
- :param selection: List of tests to be displayed in "Selected tests" window.
- :type selection: list
- :returns: List of dictionaries with "label", "value" pairs for a checkbox.
- :rtype: list
- """
- if selection:
- return [{"label": v["id"], "value": v["id"]} for v in selection]
- else:
- return list()
-
-
-def get_date(s_date: str) -> datetime:
- """Transform string reprezentation of date to datetime.datetime data type.
-
- :param s_date: String reprezentation of date.
- :type s_date: str
- :returns: Date as datetime.datetime.
- :rtype: datetime.datetime
- """
- return datetime(int(s_date[0:4]), int(s_date[5:7]), int(s_date[8:10]))
-
-
-def gen_new_url(url_components: dict, params: dict) -> str:
- """Generate a new URL with encoded parameters.
-
- :param url_components: Dictionary with URL elements. It should contain
- "scheme", "netloc" and "path".
- :param url_components: URL parameters to be encoded to the URL.
- :type parsed_url: dict
- :type params: dict
- :returns Encoded URL with parameters.
- :rtype: str
- """
-
- if url_components:
- return url_encode(
- {
- "scheme": url_components.get("scheme", ""),
- "netloc": url_components.get("netloc", ""),
- "path": url_components.get("path", ""),
- "params": params
- }
- )
- else:
- return str()
-
-
-def get_duts(df: pd.DataFrame) -> list:
- """Get the list of DUTs from the pre-processed information about jobs.
-
- :param df: DataFrame with information about jobs.
- :type df: pandas.DataFrame
- :returns: Alphabeticaly sorted list of DUTs.
- :rtype: list
- """
- return sorted(list(df["dut"].unique()))
-
-
-def get_ttypes(df: pd.DataFrame, dut: str) -> list:
- """Get the list of test types from the pre-processed information about
- jobs.
-
- :param df: DataFrame with information about jobs.
- :param dut: The DUT for which the list of test types will be populated.
- :type df: pandas.DataFrame
- :type dut: str
- :returns: Alphabeticaly sorted list of test types.
- :rtype: list
- """
- return sorted(list(df.loc[(df["dut"] == dut)]["ttype"].unique()))
-
-
-def get_cadences(df: pd.DataFrame, dut: str, ttype: str) -> list:
- """Get the list of cadences from the pre-processed information about
- jobs.
-
- :param df: DataFrame with information about jobs.
- :param dut: The DUT for which the list of cadences will be populated.
- :param ttype: The test type for which the list of cadences will be
- populated.
- :type df: pandas.DataFrame
- :type dut: str
- :type ttype: str
- :returns: Alphabeticaly sorted list of cadences.
- :rtype: list
- """
- return sorted(list(df.loc[(
- (df["dut"] == dut) &
- (df["ttype"] == ttype)
- )]["cadence"].unique()))
-
-
-def get_test_beds(df: pd.DataFrame, dut: str, ttype: str, cadence: str) -> list:
- """Get the list of test beds from the pre-processed information about
- jobs.
-
- :param df: DataFrame with information about jobs.
- :param dut: The DUT for which the list of test beds will be populated.
- :param ttype: The test type for which the list of test beds will be
- populated.
- :param cadence: The cadence for which the list of test beds will be
- populated.
- :type df: pandas.DataFrame
- :type dut: str
- :type ttype: str
- :type cadence: str
- :returns: Alphabeticaly sorted list of test beds.
- :rtype: list
- """
- return sorted(list(df.loc[(
- (df["dut"] == dut) &
- (df["ttype"] == ttype) &
- (df["cadence"] == cadence)
- )]["tbed"].unique()))
-
-
-def get_job(df: pd.DataFrame, dut, ttype, cadence, testbed):
- """Get the name of a job defined by dut, ttype, cadence, test bed.
- Input information comes from the control panel.
-
- :param df: DataFrame with information about jobs.
- :param dut: The DUT for which the job name will be created.
- :param ttype: The test type for which the job name will be created.
- :param cadence: The cadence for which the job name will be created.
- :param testbed: The test bed for which the job name will be created.
- :type df: pandas.DataFrame
- :type dut: str
- :type ttype: str
- :type cadence: str
- :type testbed: str
- :returns: Job name.
- :rtype: str
- """
- return df.loc[(
- (df["dut"] == dut) &
- (df["ttype"] == ttype) &
- (df["cadence"] == cadence) &
- (df["tbed"] == testbed)
- )]["job"].item()
-
-
-def generate_options(opts: list) -> list:
- """Return list of options for radio items in control panel. The items in
- the list are dictionaries with keys "label" and "value".
-
- :params opts: List of options (str) to be used for the generated list.
- :type opts: list
- :returns: List of options (dict).
- :rtype: list
- """
- return [{"label": i, "value": i} for i in opts]
-
-
-def set_job_params(df: pd.DataFrame, job: str) -> dict:
- """Create a dictionary with all options and values for (and from) the
- given job.
-
- :param df: DataFrame with information about jobs.
- :params job: The name of job for and from which the dictionary will be
- created.
- :type df: pandas.DataFrame
- :type job: str
- :returns: Dictionary with all options and values for (and from) the
- given job.
- :rtype: dict
- """
-
- l_job = job.split("-")
- return {
- "job": job,
- "dut": l_job[1],
- "ttype": l_job[3],
- "cadence": l_job[4],
- "tbed": "-".join(l_job[-2:]),
- "duts": generate_options(get_duts(df)),
- "ttypes": generate_options(get_ttypes(df, l_job[1])),
- "cadences": generate_options(get_cadences(df, l_job[1], l_job[3])),
- "tbeds": generate_options(
- get_test_beds(df, l_job[1], l_job[3], l_job[4]))
- }