1
0
forked from ScoDoc/ScoDoc
ScoDoc/app/scodoc/sco_report.py

1593 lines
53 KiB
Python
Raw Normal View History

2020-09-26 16:19:37 +02:00
# -*- mode: python -*-
# -*- coding: utf-8 -*-
##############################################################################
#
# Gestion scolarite IUT
#
2021-01-01 17:51:08 +01:00
# Copyright (c) 1999 - 2021 Emmanuel Viennet. All rights reserved.
2020-09-26 16:19:37 +02:00
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
#
# Emmanuel Viennet emmanuel.viennet@viennet.net
#
##############################################################################
"""Rapports suivi:
- statistiques decisions
- suivi cohortes
"""
2021-01-10 18:05:20 +01:00
import os
import tempfile
2021-07-09 17:47:06 +02:00
import six.moves.urllib.request, six.moves.urllib.parse, six.moves.urllib.error
2021-01-10 18:05:20 +01:00
import re
import time
2021-02-05 18:21:34 +01:00
import datetime
2021-07-09 23:19:30 +02:00
from operator import itemgetter
2020-09-26 16:19:37 +02:00
from flask import url_for, g, request
import pydot
import app.scodoc.sco_utils as scu
2021-08-12 14:31:15 +02:00
from app.models import NotesFormModalite
from app.scodoc import notesdb as ndb
from app.scodoc import html_sco_header
from app.scodoc import sco_codes_parcours
2021-07-19 19:53:01 +02:00
from app.scodoc import sco_cache
from app.scodoc import sco_etud
from app.scodoc import sco_excel
from app.scodoc import sco_formsemestre
from app.scodoc import sco_formsemestre_inscriptions
from app.scodoc import sco_formsemestre_status
from app.scodoc import sco_parcours_dut
from app.scodoc import sco_pdf
from app.scodoc import sco_preferences
2021-08-21 17:07:44 +02:00
import sco_version
from app.scodoc.gen_tables import GenTable
2021-08-29 19:57:32 +02:00
from app import log
from app.scodoc.sco_codes_parcours import code_semestre_validant
from app.scodoc.sco_exceptions import ScoValueError
from app.scodoc.sco_pdf import SU
2020-09-26 16:19:37 +02:00
MAX_ETUD_IN_DESCR = 20
def formsemestre_etuds_stats(sem, only_primo=False):
"""Récupère liste d'etudiants avec etat et decision."""
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(
sem["formsemestre_id"]
2020-09-26 16:19:37 +02:00
) # > get_table_moyennes_triees, identdict, get_etud_decision_sem, get_etud_etat,
T = nt.get_table_moyennes_triees()
# Construit liste d'étudiants du semestre avec leur decision
etuds = []
for t in T:
etudid = t[-1]
etud = sco_etud.get_etud_info(etudid=etudid, filled=True)[0]
2020-09-26 16:19:37 +02:00
decision = nt.get_etud_decision_sem(etudid)
if decision:
etud["codedecision"] = decision["code"]
etud["etat"] = nt.get_etud_etat(etudid)
if etud["etat"] == "D":
etud["codedecision"] = "DEM"
2021-07-09 17:47:06 +02:00
if "codedecision" not in etud:
2020-09-26 16:19:37 +02:00
etud["codedecision"] = "(nd)" # pas de decision jury
# Ajout devenir (autorisations inscriptions), utile pour stats passage
aut_list = sco_parcours_dut.formsemestre_get_autorisation_inscription(
etudid, sem["formsemestre_id"]
2020-09-26 16:19:37 +02:00
)
autorisations = ["S%s" % x["semestre_id"] for x in aut_list]
autorisations.sort()
autorisations_str = ", ".join(autorisations)
etud["devenir"] = autorisations_str
# Ajout clé 'bac-specialite'
bs = []
if etud["bac"]:
bs.append(etud["bac"])
if etud["specialite"]:
bs.append(etud["specialite"])
etud["bac-specialite"] = " ".join(bs)
#
if (not only_primo) or is_primo_etud(etud, sem):
2020-09-26 16:19:37 +02:00
etuds.append(etud)
return etuds
def is_primo_etud(etud, sem):
"""Determine si un (filled) etud a ete inscrit avant ce semestre.
Regarde la liste des semestres dans lesquels l'étudiant est inscrit
"""
now = sem["dateord"]
for s in etud["sems"]: # le + recent d'abord
if s["dateord"] < now:
return False
return True
2020-09-26 16:19:37 +02:00
def _categories_and_results(etuds, category, result):
categories = {}
results = {}
for etud in etuds:
categories[etud[category]] = True
results[etud[result]] = True
2021-07-09 17:47:06 +02:00
categories = list(categories.keys())
2020-09-26 16:19:37 +02:00
categories.sort()
2021-07-09 17:47:06 +02:00
results = list(results.keys())
2020-09-26 16:19:37 +02:00
results.sort()
return categories, results
def _results_by_category(
etuds,
category="",
result="",
category_name=None,
formsemestre_id=None,
):
"""Construit table: categories (eg types de bacs) en ligne, décisions jury en colonnes
etuds est une liste d'etuds (dicts).
category et result sont des clés de etud (category définie les lignes, result les colonnes).
Retourne une table.
"""
if category_name is None:
category_name = category
# types de bacs differents:
categories, results = _categories_and_results(etuds, category, result)
#
Count = {} # { bac : { decision : nb_avec_ce_bac_et_ce_code } }
results = {} # { result_value : True }
for etud in etuds:
results[etud[result]] = True
2021-07-09 17:47:06 +02:00
if etud[category] in Count:
2020-09-26 16:19:37 +02:00
Count[etud[category]][etud[result]] += 1
else:
2021-01-10 18:05:20 +01:00
Count[etud[category]] = scu.DictDefault(kv_dict={etud[result]: 1})
2020-09-26 16:19:37 +02:00
# conversion en liste de dict
C = [Count[cat] for cat in categories]
# Totaux par lignes et colonnes
tot = 0
for l in [Count[cat] for cat in categories]:
l["sum"] = sum(l.values())
tot += l["sum"]
# pourcentages sur chaque total de ligne
for l in C:
l["sumpercent"] = "%2.1f%%" % ((100.0 * l["sum"]) / tot)
#
2021-07-09 17:47:06 +02:00
codes = list(results.keys())
2020-09-26 16:19:37 +02:00
codes.sort()
bottom_titles = []
if C: # ligne du bas avec totaux:
bottom_titles = {}
for code in codes:
bottom_titles[code] = sum([l[code] for l in C])
bottom_titles["sum"] = tot
bottom_titles["sumpercent"] = "100%"
bottom_titles["row_title"] = "Total"
# ajout titre ligne:
for (cat, l) in zip(categories, C):
l["row_title"] = cat or "?"
#
codes.append("sum")
codes.append("sumpercent")
# on veut { ADM : ADM, ... }, était peu elegant en python 2.3:
# titles = {}
# map( lambda x,titles=titles: titles.__setitem__(x[0],x[1]), zip(codes,codes) )
# Version moderne:
titles = {x: x for x in codes}
titles["sum"] = "Total"
titles["sumpercent"] = "%"
titles["DEM"] = "Dém." # démissions
titles["row_title"] = category_name
return GenTable(
titles=titles,
columns_ids=codes,
rows=C,
bottom_titles=bottom_titles,
html_col_width="4em",
html_sortable=True,
preferences=sco_preferences.SemPreferences(formsemestre_id),
2020-09-26 16:19:37 +02:00
)
# pages
def formsemestre_report(
formsemestre_id,
etuds,
REQUEST=None,
category="bac",
result="codedecision",
category_name="",
result_name="",
title="Statistiques",
only_primo=None,
):
"""
Tableau sur résultats (result) par type de category bac
"""
2021-08-19 10:28:35 +02:00
sem = sco_formsemestre.get_formsemestre(formsemestre_id)
2020-09-26 16:19:37 +02:00
if not category_name:
category_name = category
if not result_name:
result_name = result
if result_name == "codedecision":
result_name = "résultats"
#
tab = _results_by_category(
etuds,
category=category,
category_name=category_name,
result=result,
formsemestre_id=formsemestre_id,
)
#
2021-01-10 18:05:20 +01:00
tab.filename = scu.make_filename("stats " + sem["titreannee"])
2020-09-26 16:19:37 +02:00
2021-08-21 17:07:44 +02:00
tab.origin = (
"Généré par %s le " % sco_version.SCONAME + scu.timedate_human_repr() + ""
)
2020-09-26 16:19:37 +02:00
tab.caption = "Répartition des résultats par %s, semestre %s" % (
category_name,
sem["titreannee"],
)
tab.html_caption = "Répartition des résultats par %s." % category_name
tab.base_url = "%s?formsemestre_id=%s" % (request.base_url, formsemestre_id)
2020-09-26 16:19:37 +02:00
if only_primo:
2021-05-11 11:48:32 +02:00
tab.base_url += "&only_primo=on"
2020-09-26 16:19:37 +02:00
return tab
# def formsemestre_report_bacs(formsemestre_id, format='html', REQUEST=None):
2020-09-26 16:19:37 +02:00
# """
# Tableau sur résultats par type de bac
# """
2021-08-19 10:28:35 +02:00
# sem = sco_formsemestre.get_formsemestre( formsemestre_id)
2020-09-26 16:19:37 +02:00
# title = 'Statistiques bacs ' + sem['titreannee']
# etuds = formsemestre_etuds_stats(sem)
# tab = formsemestre_report(formsemestre_id, etuds, REQUEST=REQUEST,
2020-09-26 16:19:37 +02:00
# category='bac', result='codedecision',
# category_name='Bac',
# title=title)
# return tab.make_page(
# title = """<h2>Résultats de <a href="formsemestre_status?formsemestre_id=%(formsemestre_id)s">%(titreannee)s</a></h2>""" % sem,
# format=format, page_title = title, REQUEST=REQUEST )
def formsemestre_report_counts(
formsemestre_id,
format="html",
REQUEST=None,
category="bac",
result="codedecision",
allkeys=False,
only_primo=False,
):
"""
Tableau comptage avec choix des categories
"""
2021-08-19 10:28:35 +02:00
sem = sco_formsemestre.get_formsemestre(formsemestre_id)
category_name = category.capitalize()
2020-09-26 16:19:37 +02:00
title = "Comptages " + category_name
etuds = formsemestre_etuds_stats(sem, only_primo=only_primo)
2020-09-26 16:19:37 +02:00
tab = formsemestre_report(
formsemestre_id,
etuds,
REQUEST=REQUEST,
category=category,
result=result,
category_name=category_name,
title=title,
only_primo=only_primo,
)
if not etuds:
F = ["""<p><em>Aucun étudiant</em></p>"""]
else:
if allkeys:
2021-07-09 17:47:06 +02:00
keys = list(etuds[0].keys())
2020-09-26 16:19:37 +02:00
else:
# clés présentées à l'utilisateur:
keys = [
"annee_bac",
"annee_naissance",
"bac",
"specialite",
"bac-specialite",
"codedecision",
"devenir",
"etat",
"civilite",
2020-09-26 16:19:37 +02:00
"qualite",
"villelycee",
"statut",
"type_admission",
"boursier_prec",
]
keys.sort()
F = [
"""<form name="f" method="get" action="%s"><p>
Colonnes: <select name="result" onchange="document.f.submit()">"""
% request.base_url
2020-09-26 16:19:37 +02:00
]
for k in keys:
if k == result:
selected = "selected"
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (k, selected, k))
F.append("</select>")
F.append(' Lignes: <select name="category" onchange="document.f.submit()">')
for k in keys:
if k == category:
selected = "selected"
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (k, selected, k))
F.append("</select>")
if only_primo:
checked = 'checked="1"'
else:
checked = ""
F.append(
'<br/><input type="checkbox" name="only_primo" onchange="document.f.submit()" %s>Restreindre aux primo-entrants</input>'
% checked
)
F.append(
'<input type="hidden" name="formsemestre_id" value="%s"/>' % formsemestre_id
)
F.append("</p></form>")
t = tab.make_page(
title="""<h2 class="formsemestre">Comptes croisés</h2>""",
format=format,
with_html_headers=False,
)
if format != "html":
return t
H = [
html_sco_header.sco_header(page_title=title),
2020-09-26 16:19:37 +02:00
t,
"\n".join(F),
"""<p class="help">Le tableau affiche le nombre d'étudiants de ce semestre dans chacun
des cas choisis: à l'aide des deux menus, vous pouvez choisir les catégories utilisées
pour les lignes et les colonnes. Le <tt>codedecision</tt> est le code de la décision
du jury.
</p>""",
html_sco_header.sco_footer(),
2020-09-26 16:19:37 +02:00
]
return "\n".join(H)
# --------------------------------------------------------------------------
def table_suivi_cohorte(
formsemestre_id,
percent=False,
bac="", # selection sur type de bac
bacspecialite="",
annee_bac="",
civilite=None,
2020-09-26 16:19:37 +02:00
statut="",
only_primo=False,
):
"""
Tableau indiquant le nombre d'etudiants de la cohorte dans chaque état:
2020-09-26 16:19:37 +02:00
Etat date_debut_Sn date1 date2 ...
S_n #inscrits en Sn
S_n+1
...
S_last
Diplome
Sorties
Determination des dates: on regroupe les semestres commençant à des dates proches
"""
sem = sco_formsemestre.get_formsemestre(
2021-08-19 10:28:35 +02:00
formsemestre_id
2020-09-26 16:19:37 +02:00
) # sem est le semestre origine
t0 = time.time()
def logt(op):
if 0: # debug, set to 0 in production
log("%s: %s" % (op, time.time() - t0))
logt("table_suivi_cohorte: start")
# 1-- Liste des semestres posterieurs dans lesquels ont été les etudiants de sem
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(
formsemestre_id
2020-09-26 16:19:37 +02:00
) # > get_etudids, get_etud_decision_sem
etudids = nt.get_etudids()
logt("A: orig etuds set")
S = {formsemestre_id: sem} # ensemble de formsemestre_id
orig_set = set() # ensemble d'etudid du semestre d'origine
bacs = set()
bacspecialites = set()
annee_bacs = set()
civilites = set()
statuts = set()
2020-09-26 16:19:37 +02:00
for etudid in etudids:
etud = sco_etud.get_etud_info(etudid=etudid, filled=True)[0]
2020-09-26 16:19:37 +02:00
bacspe = etud["bac"] + " / " + etud["specialite"]
# sélection sur bac:
if (
(not bac or (bac == etud["bac"]))
and (not bacspecialite or (bacspecialite == bacspe))
and (not annee_bac or (annee_bac == str(etud["annee_bac"])))
and (not civilite or (civilite == etud["civilite"]))
2020-09-26 16:19:37 +02:00
and (not statut or (statut == etud["statut"]))
and (not only_primo or is_primo_etud(etud, sem))
2020-09-26 16:19:37 +02:00
):
orig_set.add(etudid)
# semestres suivants:
for s in etud["sems"]:
if ndb.DateDMYtoISO(s["date_debut"]) > ndb.DateDMYtoISO(
sem["date_debut"]
):
2020-09-26 16:19:37 +02:00
S[s["formsemestre_id"]] = s
bacs.add(etud["bac"])
bacspecialites.add(bacspe)
2021-07-13 16:39:41 +02:00
annee_bacs.add(str(etud["annee_bac"]))
civilites.add(etud["civilite"])
2020-09-26 16:19:37 +02:00
if etud["statut"]: # ne montre pas les statuts non renseignés
statuts.add(etud["statut"])
2021-07-09 17:47:06 +02:00
sems = list(S.values())
2020-09-26 16:19:37 +02:00
# tri les semestres par date de debut
for s in sems:
d, m, y = [int(x) for x in s["date_debut"].split("/")]
2021-02-05 18:21:34 +01:00
s["date_debut_dt"] = datetime.datetime(y, m, d)
2021-07-09 23:19:30 +02:00
sems.sort(key=itemgetter("date_debut_dt"))
2020-09-26 16:19:37 +02:00
# 2-- Pour chaque semestre, trouve l'ensemble des etudiants venant de sem
logt("B: etuds sets")
sem["members"] = orig_set
for s in sems:
ins = sco_formsemestre_inscriptions.do_formsemestre_inscription_list(
2021-08-19 10:28:35 +02:00
args={"formsemestre_id": s["formsemestre_id"]}
) # sans dems
inset = set([i["etudid"] for i in ins])
2020-09-26 16:19:37 +02:00
s["members"] = orig_set.intersection(inset)
nb_dipl = 0 # combien de diplomes dans ce semestre ?
if s["semestre_id"] == nt.parcours.NB_SEM:
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(
s["formsemestre_id"]
2020-09-26 16:19:37 +02:00
) # > get_etud_decision_sem
for etudid in s["members"]:
dec = nt.get_etud_decision_sem(etudid)
if dec and code_semestre_validant(dec["code"]):
nb_dipl += 1
s["nb_dipl"] = nb_dipl
# 3-- Regroupe les semestres par date de debut
P = [] # liste de periodsem
2021-07-09 23:31:16 +02:00
class periodsem(object):
2020-09-26 16:19:37 +02:00
pass
# semestre de depart:
porigin = periodsem()
d, m, y = [int(x) for x in sem["date_debut"].split("/")]
2021-02-05 18:21:34 +01:00
porigin.datedebut = datetime.datetime(y, m, d)
2020-09-26 16:19:37 +02:00
porigin.sems = [sem]
#
2021-02-05 18:21:34 +01:00
tolerance = datetime.timedelta(days=45)
2020-09-26 16:19:37 +02:00
for s in sems:
merged = False
for p in P:
2021-02-05 18:21:34 +01:00
if abs(s["date_debut_dt"] - p.datedebut) < tolerance:
2020-09-26 16:19:37 +02:00
p.sems.append(s)
merged = True
break
if not merged:
p = periodsem()
2021-02-05 18:21:34 +01:00
p.datedebut = s["date_debut_dt"]
2020-09-26 16:19:37 +02:00
p.sems = [s]
P.append(p)
# 4-- regroupe par indice de semestre S_i
indices_sems = list(set([s["semestre_id"] for s in sems]))
2020-09-26 16:19:37 +02:00
indices_sems.sort()
for p in P:
p.nb_etuds = 0 # nombre total d'etudiants dans la periode
2021-01-10 18:05:20 +01:00
p.sems_by_id = scu.DictDefault(defaultvalue=[])
2020-09-26 16:19:37 +02:00
for s in p.sems:
p.sems_by_id[s["semestre_id"]].append(s)
p.nb_etuds += len(s["members"])
# 5-- Contruit table
logt("C: build table")
nb_initial = len(sem["members"])
def fmtval(x):
if not x:
return "" # ne montre pas les 0
if percent:
return "%2.1f%%" % (100.0 * x / nb_initial)
else:
return x
L = [
{
"row_title": "Origine: S%s" % sem["semestre_id"],
porigin.datedebut: nb_initial,
"_css_row_class": "sorttop",
}
]
if nb_initial <= MAX_ETUD_IN_DESCR:
etud_descr = _descr_etud_set(sem["members"])
2020-09-26 16:19:37 +02:00
L[0]["_%s_help" % porigin.datedebut] = etud_descr
for idx_sem in indices_sems:
if idx_sem >= 0:
d = {"row_title": "S%s" % idx_sem}
else:
d = {"row_title": "Autre semestre"}
for p in P:
etuds_period = set()
2020-09-26 16:19:37 +02:00
for s in p.sems:
if s["semestre_id"] == idx_sem:
etuds_period = etuds_period.union(s["members"])
nbetuds = len(etuds_period)
if nbetuds:
d[p.datedebut] = fmtval(nbetuds)
if nbetuds <= MAX_ETUD_IN_DESCR: # si peu d'etudiants, indique la liste
etud_descr = _descr_etud_set(etuds_period)
2020-09-26 16:19:37 +02:00
d["_%s_help" % p.datedebut] = etud_descr
L.append(d)
# Compte nb de démissions et de ré-orientation par période
logt("D: cout dems reos")
sem["dems"], sem["reos"] = _count_dem_reo(formsemestre_id, sem["members"])
2020-09-26 16:19:37 +02:00
for p in P:
p.dems = set()
p.reos = set()
2020-09-26 16:19:37 +02:00
for s in p.sems:
d, r = _count_dem_reo(s["formsemestre_id"], s["members"])
2020-09-26 16:19:37 +02:00
p.dems.update(d)
p.reos.update(r)
# Nombre total d'etudiants par periode
l = {
"row_title": "Inscrits",
"row_title_help": "Nombre d'étudiants inscrits",
"_table_part": "foot",
porigin.datedebut: fmtval(nb_initial),
}
for p in P:
l[p.datedebut] = fmtval(p.nb_etuds)
L.append(l)
# Nombre de démissions par période
l = {
"row_title": "Démissions",
"row_title_help": "Nombre de démissions pendant la période",
"_table_part": "foot",
porigin.datedebut: fmtval(len(sem["dems"])),
}
if len(sem["dems"]) <= MAX_ETUD_IN_DESCR:
etud_descr = _descr_etud_set(sem["dems"])
2020-09-26 16:19:37 +02:00
l["_%s_help" % porigin.datedebut] = etud_descr
for p in P:
l[p.datedebut] = fmtval(len(p.dems))
if len(p.dems) <= MAX_ETUD_IN_DESCR:
etud_descr = _descr_etud_set(p.dems)
2020-09-26 16:19:37 +02:00
l["_%s_help" % p.datedebut] = etud_descr
L.append(l)
# Nombre de réorientations par période
l = {
"row_title": "Echecs",
"row_title_help": "Ré-orientations (décisions NAR)",
"_table_part": "foot",
porigin.datedebut: fmtval(len(sem["reos"])),
}
if len(sem["reos"]) < 10:
etud_descr = _descr_etud_set(sem["reos"])
2020-09-26 16:19:37 +02:00
l["_%s_help" % porigin.datedebut] = etud_descr
for p in P:
l[p.datedebut] = fmtval(len(p.reos))
if len(p.reos) <= MAX_ETUD_IN_DESCR:
etud_descr = _descr_etud_set(p.reos)
2020-09-26 16:19:37 +02:00
l["_%s_help" % p.datedebut] = etud_descr
L.append(l)
# derniere ligne: nombre et pourcentage de diplomes
l = {
"row_title": "Diplômes",
"row_title_help": "Nombre de diplômés à la fin de la période",
"_table_part": "foot",
}
for p in P:
nb_dipl = 0
for s in p.sems:
nb_dipl += s["nb_dipl"]
l[p.datedebut] = fmtval(nb_dipl)
L.append(l)
columns_ids = [p.datedebut for p in P]
titles = dict([(p.datedebut, p.datedebut.strftime("%d/%m/%y")) for p in P])
titles[porigin.datedebut] = porigin.datedebut.strftime("%d/%m/%y")
if percent:
pp = "(en % de la population initiale) "
titles["row_title"] = "%"
else:
pp = ""
titles["row_title"] = ""
if only_primo:
pp += "(restreint aux primo-entrants) "
if bac:
dbac = " (bacs %s)" % bac
else:
dbac = ""
if bacspecialite:
dbac += " (spécialité %s)" % bacspecialite
if annee_bac:
dbac += " (année bac %s)" % annee_bac
if civilite:
dbac += " civilité: %s" % civilite
2020-09-26 16:19:37 +02:00
if statut:
dbac += " statut: %s" % statut
tab = GenTable(
titles=titles,
columns_ids=columns_ids,
rows=L,
html_col_width="4em",
html_sortable=True,
2021-01-10 18:05:20 +01:00
filename=scu.make_filename("cohorte " + sem["titreannee"]),
2021-08-21 17:07:44 +02:00
origin="Généré par %s le " % sco_version.SCONAME
+ scu.timedate_human_repr()
+ "",
2020-09-26 16:19:37 +02:00
caption="Suivi cohorte " + pp + sem["titreannee"] + dbac,
page_title="Suivi cohorte " + sem["titreannee"],
html_class="table_cohorte",
preferences=sco_preferences.SemPreferences(formsemestre_id),
2020-09-26 16:19:37 +02:00
)
# Explication: liste des semestres associés à chaque date
if not P:
expl = [
'<p class="help">(aucun étudiant trouvé dans un semestre ultérieur)</p>'
]
else:
expl = ["<h3>Semestres associés à chaque date:</h3><ul>"]
for p in P:
expl.append("<li><b>%s</b>:" % p.datedebut.strftime("%d/%m/%y"))
ls = []
for s in p.sems:
ls.append(
'<a href="formsemestre_status?formsemestre_id=%(formsemestre_id)s">%(titreannee)s</a>'
% s
)
expl.append(", ".join(ls) + "</li>")
expl.append("</ul>")
logt("Z: table_suivi_cohorte done")
return (
tab,
"\n".join(expl),
bacs,
bacspecialites,
annee_bacs,
civilites,
statuts,
)
2020-09-26 16:19:37 +02:00
def formsemestre_suivi_cohorte(
formsemestre_id,
format="html",
percent=1,
bac="",
bacspecialite="",
annee_bac="",
civilite=None,
2020-09-26 16:19:37 +02:00
statut="",
only_primo=False,
REQUEST=None,
):
"""Affiche suivi cohortes par numero de semestre"""
2021-08-21 17:37:38 +02:00
annee_bac = str(annee_bac)
2020-09-26 16:19:37 +02:00
percent = int(percent)
(
tab,
expl,
bacs,
bacspecialites,
annee_bacs,
civilites,
statuts,
) = table_suivi_cohorte(
2020-09-26 16:19:37 +02:00
formsemestre_id,
percent=percent,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
only_primo=only_primo,
)
tab.base_url = (
2021-05-11 11:48:32 +02:00
"%s?formsemestre_id=%s&percent=%s&bac=%s&bacspecialite=%s&civilite=%s"
% (request.base_url, formsemestre_id, percent, bac, bacspecialite, civilite)
2020-09-26 16:19:37 +02:00
)
if only_primo:
2021-05-11 11:48:32 +02:00
tab.base_url += "&only_primo=on"
t = tab.make_page(format=format, with_html_headers=False)
2020-09-26 16:19:37 +02:00
if format != "html":
return t
base_url = request.base_url
burl = "%s?formsemestre_id=%s&bac=%s&bacspecialite=%s&civilite=%s&statut=%s" % (
base_url,
formsemestre_id,
bac,
bacspecialite,
civilite,
statut,
2020-09-26 16:19:37 +02:00
)
if percent:
pplink = '<p><a href="%s&percent=0">Afficher les résultats bruts</a></p>' % burl
2020-09-26 16:19:37 +02:00
else:
pplink = (
2021-05-11 11:48:32 +02:00
'<p><a href="%s&percent=1">Afficher les résultats en pourcentages</a></p>'
2020-09-26 16:19:37 +02:00
% burl
)
help = (
pplink
+ """
<p class="help">Nombre d'étudiants dans chaque semestre. Les dates indiquées sont les dates approximatives de <b>début</b> des semestres (les semestres commençant à des dates proches sont groupés). Le nombre de diplômés est celui à la <b>fin</b> du semestre correspondant. Lorsqu'il y a moins de %s étudiants dans une case, vous pouvez afficher leurs noms en passant le curseur sur le chiffre.</p>
<p class="help">Les menus permettent de n'étudier que certaines catégories d'étudiants (titulaires d'un type de bac, garçons ou filles). La case "restreindre aux primo-entrants" permet de ne considérer que les étudiants qui n'ont jamais été inscrits dans ScoDoc avant le semestre considéré.</p>
"""
% (MAX_ETUD_IN_DESCR,)
)
H = [
html_sco_header.sco_header(page_title=tab.page_title),
2020-09-26 16:19:37 +02:00
"""<h2 class="formsemestre">Suivi cohorte: devenir des étudiants de ce semestre</h2>""",
_gen_form_selectetuds(
formsemestre_id,
REQUEST=REQUEST,
only_primo=only_primo,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
bacs=bacs,
bacspecialites=bacspecialites,
annee_bacs=annee_bacs,
civilites=civilites,
2020-09-26 16:19:37 +02:00
statuts=statuts,
percent=percent,
),
t,
help,
expl,
html_sco_header.sco_footer(),
2020-09-26 16:19:37 +02:00
]
return "\n".join(H)
def _gen_form_selectetuds(
formsemestre_id,
REQUEST=None,
percent=None,
only_primo=None,
bac=None,
bacspecialite=None,
annee_bac=None,
civilite=None,
2020-09-26 16:19:37 +02:00
statut=None,
bacs=None,
bacspecialites=None,
annee_bacs=None,
civilites=None,
2020-09-26 16:19:37 +02:00
statuts=None,
):
"""HTML form pour choix criteres selection etudiants"""
bacs = list(bacs)
bacs.sort()
bacspecialites = list(bacspecialites)
bacspecialites.sort()
2021-08-21 17:37:38 +02:00
# on peut avoir un mix de chaines vides et d'entiers:
annee_bacs = [int(x) if x else 0 for x in annee_bacs]
annee_bacs.sort()
civilites = list(civilites)
civilites.sort()
2020-09-26 16:19:37 +02:00
statuts = list(statuts)
statuts.sort()
#
if bac:
selected = ""
else:
selected = 'selected="selected"'
F = [
"""<form id="f" method="get" action="%s">
<p>Bac: <select name="bac" onchange="javascript: submit(this);">
<option value="" %s>tous</option>
"""
% (request.base_url, selected)
2020-09-26 16:19:37 +02:00
]
for b in bacs:
if bac == b:
selected = 'selected="selected"'
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (b, selected, b))
F.append("</select>")
if bacspecialite:
selected = ""
else:
selected = 'selected="selected"'
F.append(
"""&nbsp; Bac/Specialité: <select name="bacspecialite" onchange="javascript: submit(this);">
<option value="" %s>tous</option>
"""
% selected
)
for b in bacspecialites:
if bacspecialite == b:
selected = 'selected="selected"'
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (b, selected, b))
F.append("</select>")
#
if annee_bac:
selected = ""
else:
selected = 'selected="selected"'
F.append(
"""&nbsp; Année bac: <select name="annee_bac" onchange="javascript: submit(this);">
<option value="" %s>tous</option>
"""
% selected
)
for b in annee_bacs:
if str(annee_bac) == str(b):
selected = 'selected="selected"'
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (b, selected, b))
F.append("</select>")
#
2020-09-26 16:19:37 +02:00
F.append(
"""&nbsp; Genre: <select name="civilite" onchange="javascript: submit(this);">
2020-09-26 16:19:37 +02:00
<option value="" %s>tous</option>
"""
% selected
)
for b in civilites:
if civilite == b:
2020-09-26 16:19:37 +02:00
selected = 'selected="selected"'
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (b, selected, b))
F.append("</select>")
F.append(
"""&nbsp; Statut: <select name="statut" onchange="javascript: submit(this);">
<option value="" %s>tous</option>
"""
% selected
)
for b in statuts:
if statut == b:
selected = 'selected="selected"'
else:
selected = ""
F.append('<option value="%s" %s>%s</option>' % (b, selected, b))
F.append("</select>")
if only_primo:
checked = 'checked="1"'
else:
checked = ""
F.append(
'<br/><input type="checkbox" name="only_primo" onchange="javascript: submit(this);" %s/>Restreindre aux primo-entrants'
% checked
)
F.append(
'<input type="hidden" name="formsemestre_id" value="%s"/>' % formsemestre_id
)
F.append('<input type="hidden" name="percent" value="%s"/>' % percent)
F.append("</p></form>")
return "\n".join(F)
def _descr_etud_set(etudids):
2020-09-26 16:19:37 +02:00
"textual html description of a set of etudids"
etuds = []
for etudid in etudids:
etuds.append(sco_etud.get_etud_info(etudid=etudid, filled=True)[0])
2020-09-26 16:19:37 +02:00
# sort by name
2021-07-10 13:58:25 +02:00
etuds.sort(key=itemgetter("nom"))
2020-09-26 16:19:37 +02:00
return ", ".join([e["nomprenom"] for e in etuds])
def _count_dem_reo(formsemestre_id, etudids):
2020-09-26 16:19:37 +02:00
"count nb of demissions and reorientation in this etud set"
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(
formsemestre_id
2020-09-26 16:19:37 +02:00
) # > get_etud_etat, get_etud_decision_sem
dems = set()
reos = set()
2020-09-26 16:19:37 +02:00
for etudid in etudids:
if nt.get_etud_etat(etudid) == "D":
dems.add(etudid)
dec = nt.get_etud_decision_sem(etudid)
if dec and dec["code"] in sco_codes_parcours.CODES_SEM_REO:
reos.add(etudid)
return dems, reos
"""OLDGEA:
27s pour S1 F.I. classique Semestre 1 2006-2007
B 2.3s
C 5.6s
D 5.9s
Z 27s => cache des semestres pour nt
à chaud: 3s
B: etuds sets: 2.4s => lent: N x getEtudInfo (non caché)
"""
EXP_LIC = re.compile(r"licence", re.I)
EXP_LPRO = re.compile(r"professionnelle", re.I)
def _codesem(sem, short=True, prefix=""):
"code semestre: S1 ou S1d"
idx = sem["semestre_id"]
# semestre décalé ?
# les semestres pairs normaux commencent entre janvier et mars
# les impairs normaux entre aout et decembre
d = ""
if idx and idx > 0 and sem["date_debut"]:
mois_debut = int(sem["date_debut"].split("/")[1])
if (idx % 2 and mois_debut < 3) or (idx % 2 == 0 and mois_debut >= 8):
d = "d"
if idx == -1:
if short:
idx = "Autre "
else:
idx = sem["titre"] + " "
idx = EXP_LPRO.sub("pro.", idx)
idx = EXP_LIC.sub("Lic.", idx)
prefix = "" # indique titre au lieu de Sn
return "%s%s%s" % (prefix, idx, d)
def get_codeparcoursetud(etud, prefix="", separator=""):
2020-09-26 16:19:37 +02:00
"""calcule un code de parcours pour un etudiant
exemples:
1234A pour un etudiant ayant effectué S1, S2, S3, S4 puis diplome
12D pour un étudiant en S1, S2 puis démission en S2
12R pour un etudiant en S1, S2 réorienté en fin de S2
Construit aussi un dict: { semestre_id : decision_jury | None }
2020-09-26 16:19:37 +02:00
"""
p = []
decisions_jury = {}
# élimine les semestres spéciaux sans parcours (LP...)
sems = [s for s in etud["sems"] if s["semestre_id"] >= 0]
i = len(sems) - 1
while i >= 0:
s = sems[i] # 'sems' est a l'envers, du plus recent au plus ancien
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(
s["formsemestre_id"]
2020-09-26 16:19:37 +02:00
) # > get_etud_etat, get_etud_decision_sem
p.append(_codesem(s, prefix=prefix))
# code decisions jury de chaque semestre:
if nt.get_etud_etat(etud["etudid"]) == "D":
decisions_jury[s["semestre_id"]] = "DEM"
else:
dec = nt.get_etud_decision_sem(etud["etudid"])
if not dec:
decisions_jury[s["semestre_id"]] = ""
else:
decisions_jury[s["semestre_id"]] = dec["code"]
# code etat dans le codeparcours sur dernier semestre seulement
if i == 0:
# Démission
if nt.get_etud_etat(etud["etudid"]) == "D":
p.append(":D")
else:
dec = nt.get_etud_decision_sem(etud["etudid"])
if dec and dec["code"] in sco_codes_parcours.CODES_SEM_REO:
p.append(":R")
if (
dec
and s["semestre_id"] == nt.parcours.NB_SEM
and code_semestre_validant(dec["code"])
):
p.append(":A")
i -= 1
return separator.join(p), decisions_jury
def tsp_etud_list(
formsemestre_id,
only_primo=False,
bac="", # selection sur type de bac
bacspecialite="",
annee_bac="",
civilite="",
2020-09-26 16:19:37 +02:00
statut="",
):
"""Liste des etuds a considerer dans table suivi parcours
ramene aussi ensembles des bacs, genres, statuts de (tous) les etudiants
2020-09-26 16:19:37 +02:00
"""
# log('tsp_etud_list(%s, bac="%s")' % (formsemestre_id,bac))
2021-08-19 10:28:35 +02:00
sem = sco_formsemestre.get_formsemestre(formsemestre_id)
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(formsemestre_id) # > get_etudids,
2020-09-26 16:19:37 +02:00
etudids = nt.get_etudids()
etuds = []
bacs = set()
bacspecialites = set()
annee_bacs = set()
civilites = set()
statuts = set()
2020-09-26 16:19:37 +02:00
for etudid in etudids:
etud = sco_etud.get_etud_info(etudid=etudid, filled=True)[0]
2020-09-26 16:19:37 +02:00
bacspe = etud["bac"] + " / " + etud["specialite"]
# sélection sur bac, primo, ...:
if (
(not bac or (bac == etud["bac"]))
and (not bacspecialite or (bacspecialite == bacspe))
and (not annee_bac or (annee_bac == str(etud["annee_bac"])))
and (not civilite or (civilite == etud["civilite"]))
2020-09-26 16:19:37 +02:00
and (not statut or (statut == etud["statut"]))
and (not only_primo or is_primo_etud(etud, sem))
2020-09-26 16:19:37 +02:00
):
etuds.append(etud)
bacs.add(etud["bac"])
bacspecialites.add(bacspe)
annee_bacs.add(etud["annee_bac"])
civilites.add(etud["civilite"])
2020-09-26 16:19:37 +02:00
if etud["statut"]: # ne montre pas les statuts non renseignés
statuts.add(etud["statut"])
# log('tsp_etud_list: %s etuds' % len(etuds))
return etuds, bacs, bacspecialites, annee_bacs, civilites, statuts
2020-09-26 16:19:37 +02:00
def tsp_grouped_list(codes_etuds):
"""Liste pour table regroupant le nombre d'étudiants (+ bulle avec les noms) de chaque parcours"""
2020-09-26 16:19:37 +02:00
L = []
2021-07-09 17:47:06 +02:00
parcours = list(codes_etuds.keys())
2020-09-26 16:19:37 +02:00
parcours.sort()
for p in parcours:
nb = len(codes_etuds[p])
l = {"parcours": p, "nb": nb}
if nb <= MAX_ETUD_IN_DESCR:
l["_nb_help"] = _descr_etud_set([e["etudid"] for e in codes_etuds[p]])
2020-09-26 16:19:37 +02:00
L.append(l)
# tri par effectifs décroissants
2021-07-09 23:19:30 +02:00
L.sort(key=itemgetter("nb"))
2020-09-26 16:19:37 +02:00
return L
def table_suivi_parcours(formsemestre_id, only_primo=False, grouped_parcours=True):
"""Tableau recapitulant tous les parcours"""
2021-08-19 10:28:35 +02:00
sem = sco_formsemestre.get_formsemestre(formsemestre_id)
etuds, bacs, bacspecialites, annee_bacs, civilites, statuts = tsp_etud_list(
formsemestre_id, only_primo=only_primo
2020-09-26 16:19:37 +02:00
)
2021-01-10 18:05:20 +01:00
codes_etuds = scu.DictDefault(defaultvalue=[])
2020-09-26 16:19:37 +02:00
for etud in etuds:
etud["codeparcours"], etud["decisions_jury"] = get_codeparcoursetud(etud)
2020-09-26 16:19:37 +02:00
codes_etuds[etud["codeparcours"]].append(etud)
fiche_url = url_for(
"scolar.ficheEtud", scodoc_dept=g.scodoc_dept, etudid=etud["etudid"]
)
etud["_nom_target"] = fiche_url
etud["_prenom_target"] = fiche_url
2020-09-26 16:19:37 +02:00
etud["_nom_td_attrs"] = 'id="%s" class="etudinfo"' % (etud["etudid"])
titles = {
"parcours": "Code parcours",
"nb": "Nombre d'étudiants",
"civilite": "",
2020-09-26 16:19:37 +02:00
"nom": "Nom",
"prenom": "Prénom",
"etudid": "etudid",
"codeparcours": "Code parcours",
"bac": "Bac",
"specialite": "Spe.",
}
if grouped_parcours:
L = tsp_grouped_list(codes_etuds)
2020-09-26 16:19:37 +02:00
columns_ids = ("parcours", "nb")
else:
# Table avec le parcours de chaque étudiant:
L = etuds
columns_ids = (
"etudid",
"civilite",
2020-09-26 16:19:37 +02:00
"nom",
"prenom",
"bac",
"specialite",
"codeparcours",
)
# Calcule intitulés de colonnes
S = set()
2021-07-09 17:47:06 +02:00
sems_ids = list(S.union(*[list(e["decisions_jury"].keys()) for e in etuds]))
2020-09-26 16:19:37 +02:00
sems_ids.sort()
sem_tits = ["S%s" % s for s in sems_ids]
titles.update([(s, s) for s in sem_tits])
columns_ids += tuple(sem_tits)
for etud in etuds:
for s in etud["decisions_jury"]:
etud["S%s" % s] = etud["decisions_jury"][s]
if only_primo:
primostr = "primo-entrants du"
else:
primostr = "passés dans le"
tab = GenTable(
columns_ids=columns_ids,
rows=L,
titles=titles,
2021-08-21 17:07:44 +02:00
origin="Généré par %s le " % sco_version.SCONAME
+ scu.timedate_human_repr()
+ "",
2020-09-26 16:19:37 +02:00
caption="Parcours suivis, étudiants %s semestre " % primostr
+ sem["titreannee"],
page_title="Parcours " + sem["titreannee"],
html_sortable=True,
html_class="table_leftalign table_listegroupe",
html_next_section="""<table class="help">
<tr><td><tt>1, 2, ...</tt></td><td> numéros de semestres</td></tr>
<tr><td><tt>1d, 2d, ...</tt></td><td>semestres "décalés"</td></tr>
<tr><td><tt>:A</tt></td><td> étudiants diplômés</td></tr>
<tr><td><tt>:R</tt></td><td> étudiants réorientés</td></tr>
<tr><td><tt>:D</tt></td><td> étudiants démissionnaires</td></tr>
</table>""",
bottom_titles={
"parcours": "Total",
"nb": len(etuds),
"codeparcours": len(etuds),
},
preferences=sco_preferences.SemPreferences(formsemestre_id),
2020-09-26 16:19:37 +02:00
)
return tab
def tsp_form_primo_group(REQUEST, only_primo, no_grouping, formsemestre_id, format):
"""Element de formulaire pour choisir si restriction aux primos entrants et groupement par lycees"""
F = ["""<form name="f" method="get" action="%s">""" % request.base_url]
2020-09-26 16:19:37 +02:00
if only_primo:
checked = 'checked="1"'
else:
checked = ""
F.append(
'<input type="checkbox" name="only_primo" onchange="document.f.submit()" %s>Restreindre aux primo-entrants</input>'
% checked
)
if no_grouping:
checked = 'checked="1"'
else:
checked = ""
F.append(
'<input type="checkbox" name="no_grouping" onchange="document.f.submit()" %s>Lister chaque étudiant</input>'
% checked
)
F.append(
'<input type="hidden" name="formsemestre_id" value="%s"/>' % formsemestre_id
)
F.append('<input type="hidden" name="format" value="%s"/>' % format)
F.append("""</form>""")
return "\n".join(F)
def formsemestre_suivi_parcours(
formsemestre_id,
format="html",
only_primo=False,
no_grouping=False,
REQUEST=None,
):
"""Effectifs dans les differents parcours possibles."""
2020-09-26 16:19:37 +02:00
tab = table_suivi_parcours(
formsemestre_id,
only_primo=only_primo,
grouped_parcours=not no_grouping,
)
tab.base_url = "%s?formsemestre_id=%s" % (request.base_url, formsemestre_id)
2020-09-26 16:19:37 +02:00
if only_primo:
2021-05-11 11:48:32 +02:00
tab.base_url += "&only_primo=1"
2020-09-26 16:19:37 +02:00
if no_grouping:
2021-05-11 11:48:32 +02:00
tab.base_url += "&no_grouping=1"
t = tab.make_page(format=format, with_html_headers=False)
2020-09-26 16:19:37 +02:00
if format != "html":
return t
F = [
tsp_form_primo_group(REQUEST, only_primo, no_grouping, formsemestre_id, format)
]
H = [
2021-06-13 19:12:20 +02:00
html_sco_header.sco_header(
2020-09-26 16:19:37 +02:00
page_title=tab.page_title,
init_qtip=True,
javascripts=["js/etud_info.js"],
),
"""<h2 class="formsemestre">Parcours suivis par les étudiants de ce semestre</h2>""",
"\n".join(F),
t,
html_sco_header.sco_footer(),
2020-09-26 16:19:37 +02:00
]
return "\n".join(H)
# -------------
def graph_parcours(
formsemestre_id,
format="svg",
only_primo=False,
bac="", # selection sur type de bac
bacspecialite="",
annee_bac="",
civilite="",
2020-09-26 16:19:37 +02:00
statut="",
):
""""""
etuds, bacs, bacspecialites, annee_bacs, civilites, statuts = tsp_etud_list(
2020-09-26 16:19:37 +02:00
formsemestre_id,
only_primo=only_primo,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
)
# log('graph_parcours: %s etuds (only_primo=%s)' % (len(etuds), only_primo))
if not etuds:
return "", etuds, bacs, bacspecialites, annee_bacs, civilites, statuts
2021-01-10 18:05:20 +01:00
edges = scu.DictDefault(
defaultvalue=set()
2021-08-12 14:31:15 +02:00
) # {("SEM"formsemestre_id_origin, "SEM"formsemestre_id_dest) : etud_set}
def sem_node_name(sem, prefix="SEM"):
"pydot node name for this integer id"
return prefix + str(sem["formsemestre_id"])
2020-09-26 16:19:37 +02:00
sems = {}
2021-01-10 18:05:20 +01:00
effectifs = scu.DictDefault(defaultvalue=set()) # formsemestre_id : etud_set
decisions = scu.DictDefault(defaultvalue={}) # formsemestre_id : { code : nb_etud }
2021-08-12 14:31:15 +02:00
isolated_nodes = [] # [ node_name_de_formsemestre_id, ... ]
connected_nodes = set() # { node_name_de_formsemestre_id }
2020-09-26 16:19:37 +02:00
diploma_nodes = []
2021-08-12 14:31:15 +02:00
dem_nodes = {} # formsemestre_id : noeud (node name) pour demissionnaires
2020-09-26 16:19:37 +02:00
nar_nodes = {} # formsemestre_id : noeud pour NAR
for etud in etuds:
2021-02-01 23:54:46 +01:00
nxt = {}
2020-09-26 16:19:37 +02:00
etudid = etud["etudid"]
for s in etud["sems"]: # du plus recent au plus ancien
2021-07-19 19:53:01 +02:00
nt = sco_cache.NotesTableCache.get(
s["formsemestre_id"]
2020-09-26 16:19:37 +02:00
) # > get_etud_decision_sem, get_etud_etat
dec = nt.get_etud_decision_sem(etudid)
if nxt:
2020-09-26 16:19:37 +02:00
if (
s["semestre_id"] == nt.parcours.NB_SEM
and dec
and code_semestre_validant(dec["code"])
and nt.get_etud_etat(etudid) == "I"
):
# cas particulier du diplome puis poursuite etude
edges[
2021-08-12 14:31:15 +02:00
(
sem_node_name(s, "_dipl_"),
sem_node_name(nxt), # = "SEM{formsemestre_id}"
)
2020-09-26 16:19:37 +02:00
].add(etudid)
else:
2021-08-12 14:31:15 +02:00
edges[(sem_node_name(s), sem_node_name(nxt))].add(etudid)
connected_nodes.add(sem_node_name(s))
connected_nodes.add(sem_node_name(nxt))
2020-09-26 16:19:37 +02:00
else:
2021-08-12 14:31:15 +02:00
isolated_nodes.append(sem_node_name(s))
2020-09-26 16:19:37 +02:00
sems[s["formsemestre_id"]] = s
effectifs[s["formsemestre_id"]].add(etudid)
nxt = s
2020-09-26 16:19:37 +02:00
# Compte decisions jury de chaque semestres:
dc = decisions[s["formsemestre_id"]]
if dec:
if dec["code"] in dc:
dc[dec["code"]] += 1
else:
dc[dec["code"]] = 1
# ajout noeud pour demissionnaires
if nt.get_etud_etat(etudid) == "D":
2021-08-12 14:31:15 +02:00
nid = sem_node_name(s, "_dem_")
2020-09-26 16:19:37 +02:00
dem_nodes[s["formsemestre_id"]] = nid
2021-08-12 14:31:15 +02:00
edges[(sem_node_name(s), nid)].add(etudid)
2020-09-26 16:19:37 +02:00
# ajout noeud pour NAR (seulement pour noeud de depart)
2021-01-10 18:05:20 +01:00
if (
s["formsemestre_id"] == formsemestre_id
and dec
and dec["code"] == sco_codes_parcours.NAR
):
2021-08-12 14:31:15 +02:00
nid = sem_node_name(s, "_nar_")
2020-09-26 16:19:37 +02:00
nar_nodes[s["formsemestre_id"]] = nid
2021-08-12 14:31:15 +02:00
edges[(sem_node_name(s), nid)].add(etudid)
2020-09-26 16:19:37 +02:00
# si "terminal", ajoute noeud pour diplomes
if s["semestre_id"] == nt.parcours.NB_SEM:
if (
dec
and code_semestre_validant(dec["code"])
and nt.get_etud_etat(etudid) == "I"
):
2021-08-12 14:31:15 +02:00
nid = sem_node_name(s, "_dipl_")
edges[(sem_node_name(s), nid)].add(etudid)
2020-09-26 16:19:37 +02:00
diploma_nodes.append(nid)
#
g = scu.graph_from_edges(list(edges.keys()))
2020-09-26 16:19:37 +02:00
for fid in isolated_nodes:
if not fid in connected_nodes:
n = pydot.Node(name=fid)
2020-09-26 16:19:37 +02:00
g.add_node(n)
g.set("rankdir", "LR") # left to right
g.set_fontname("Helvetica")
if format == "svg":
g.set_bgcolor("#fffff0") # ou 'transparent'
# titres des semestres:
for s in sems.values():
2021-08-12 14:31:15 +02:00
n = g.get_node(sem_node_name(s))[0]
2020-09-26 16:19:37 +02:00
log("s['formsemestre_id'] = %s" % s["formsemestre_id"])
log("n=%s" % n)
2021-08-12 14:31:15 +02:00
log("get=%s" % g.get_node(sem_node_name(s)))
2020-09-26 16:19:37 +02:00
log("nodes names = %s" % [x.get_name() for x in g.get_node_list()])
2021-08-12 14:31:15 +02:00
if s["modalite"] and s["modalite"] != NotesFormModalite.DEFAULT_MODALITE:
2020-09-26 16:19:37 +02:00
modalite = " " + s["modalite"]
else:
modalite = ""
label = "%s%s\\n%d/%s - %d/%s\\n%d" % (
_codesem(s, short=False, prefix="S"),
modalite,
s["mois_debut_ord"],
s["annee_debut"][2:],
s["mois_fin_ord"],
s["annee_fin"][2:],
len(effectifs[s["formsemestre_id"]]),
)
2021-01-10 18:05:20 +01:00
n.set("label", scu.suppress_accents(label))
2020-09-26 16:19:37 +02:00
n.set_fontname("Helvetica")
n.set_fontsize(8.0)
n.set_width(1.2)
n.set_shape("box")
2021-08-12 14:31:15 +02:00
n.set_URL(f"formsemestre_status?formsemestre_id={s['formsemestre_id']}")
2020-09-26 16:19:37 +02:00
# semestre de depart en vert
2021-08-12 14:31:15 +02:00
n = g.get_node("SEM" + str(formsemestre_id))[0]
2020-09-26 16:19:37 +02:00
n.set_color("green")
# demissions en rouge, octagonal
for nid in dem_nodes.values():
n = g.get_node(nid)[0]
2020-09-26 16:19:37 +02:00
n.set_color("red")
n.set_shape("octagon")
n.set("label", "Dem.")
# NAR en rouge, Mcircle
for nid in nar_nodes.values():
n = g.get_node(nid)[0]
2020-09-26 16:19:37 +02:00
n.set_color("red")
n.set_shape("Mcircle")
2021-01-10 18:05:20 +01:00
n.set("label", sco_codes_parcours.NAR)
2020-09-26 16:19:37 +02:00
# diplomes:
for nid in diploma_nodes:
n = g.get_node(nid)[0]
2020-09-26 16:19:37 +02:00
n.set_color("red")
n.set_shape("ellipse")
n.set("label", "Diplome") # bug si accent (pas compris pourquoi)
# Arètes:
bubbles = {} # substitue titres pour bulle aides: src_id:dst_id : etud_descr
for (src_id, dst_id) in edges.keys():
e = g.get_edge(src_id, dst_id)[0]
2020-09-26 16:19:37 +02:00
e.set("arrowhead", "normal")
e.set("arrowsize", 1)
e.set_label(len(edges[(src_id, dst_id)]))
e.set_fontname("Helvetica")
e.set_fontsize(8.0)
# bulle avec liste etudiants
if len(edges[(src_id, dst_id)]) <= MAX_ETUD_IN_DESCR:
etud_descr = _descr_etud_set(edges[(src_id, dst_id)])
2020-09-26 16:19:37 +02:00
bubbles[src_id + ":" + dst_id] = etud_descr
e.set_URL("__xxxetudlist__?" + src_id + ":" + dst_id)
# Genere graphe
2021-01-10 18:05:20 +01:00
_, path = tempfile.mkstemp(".gr")
2020-09-26 16:19:37 +02:00
g.write(path=path, format=format)
data = open(path, "rb").read()
2020-09-26 16:19:37 +02:00
log("dot generated %d bytes in %s format" % (len(data), format))
if not data:
log("graph.to_string=%s" % g.to_string())
raise ValueError(
"Erreur lors de la génération du document au format %s" % format
)
os.unlink(path)
if format == "svg":
# dot génère un document XML complet, il faut enlever l'en-tête
2021-07-27 13:50:53 +02:00
data_str = data.decode("utf-8")
data = "<svg" + "<svg".join(data_str.split("<svg")[1:])
2020-09-26 16:19:37 +02:00
# Substitution des titres des URL des aretes pour bulles aide
def repl(m):
return '<a title="%s"' % bubbles[m.group("sd")]
exp = re.compile(
r'<a.*?href="__xxxetudlist__\?(?P<sd>\w*:\w*).*?".*?xlink:title=".*?"', re.M
)
data = exp.sub(repl, data)
# Substitution des titres des boites (semestres)
exp1 = re.compile(
r'<a xlink:href="formsemestre_status\?formsemestre_id=(?P<fid>\w*).*?".*?xlink:title="(?P<title>.*?)"',
re.M | re.DOTALL,
)
def repl_title(m):
2021-08-12 14:31:15 +02:00
fid = int(m.group("fid"))
2020-09-26 16:19:37 +02:00
title = sems[fid]["titreannee"]
if decisions[fid]:
title += (
(" (" + str(decisions[fid]) + ")").replace("{", "").replace("'", "")
)
return (
'<a xlink:href="formsemestre_status?formsemestre_id=%s" xlink:title="%s"'
2021-01-10 18:05:20 +01:00
% (fid, scu.suppress_accents(title))
2020-09-26 16:19:37 +02:00
) # evite accents car svg utf-8 vs page en latin1...
data = exp1.sub(repl_title, data)
# Substitution de Arial par Helvetica (new prblem in Debian 5) ???
# bug turnaround: il doit bien y avoir un endroit ou regler cela ?
# cf http://groups.google.com/group/pydot/browse_thread/thread/b3704c53e331e2ec
data = data.replace("font-family:Arial", "font-family:Helvetica")
return data, etuds, bacs, bacspecialites, annee_bacs, civilites, statuts
2020-09-26 16:19:37 +02:00
def formsemestre_graph_parcours(
formsemestre_id,
format="html",
only_primo=False,
bac="", # selection sur type de bac
bacspecialite="",
annee_bac="",
civilite="",
2020-09-26 16:19:37 +02:00
statut="",
allkeys=False, # unused
2020-09-26 16:19:37 +02:00
REQUEST=None,
):
"""Graphe suivi cohortes"""
2021-08-21 17:37:38 +02:00
annee_bac = str(annee_bac)
2020-09-26 16:19:37 +02:00
# log("formsemestre_graph_parcours")
2021-08-19 10:28:35 +02:00
sem = sco_formsemestre.get_formsemestre(formsemestre_id)
2020-09-26 16:19:37 +02:00
if format == "pdf":
(
doc,
etuds,
bacs,
bacspecialites,
annee_bacs,
civilites,
statuts,
) = graph_parcours(
2020-09-26 16:19:37 +02:00
formsemestre_id,
format="pdf",
only_primo=only_primo,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
)
2021-01-10 18:05:20 +01:00
filename = scu.make_filename("flux " + sem["titreannee"])
return scu.sendPDFFile(doc, filename + ".pdf")
2020-09-26 16:19:37 +02:00
elif format == "png":
#
(
doc,
etuds,
bacs,
bacspecialites,
annee_bacs,
civilites,
statuts,
) = graph_parcours(
2020-09-26 16:19:37 +02:00
formsemestre_id,
format="png",
only_primo=only_primo,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
)
2021-01-10 18:05:20 +01:00
filename = scu.make_filename("flux " + sem["titreannee"])
2020-09-26 16:19:37 +02:00
REQUEST.RESPONSE.setHeader(
"content-disposition", 'attachment; filename="%s"' % filename
)
REQUEST.RESPONSE.setHeader("content-type", "image/png")
return doc
elif format == "html":
url_kw = {
"scodoc_dept": g.scodoc_dept,
"formsemestre_id": formsemestre_id,
"bac": bac,
"specialite": bacspecialite,
"civilite": civilite,
"statut": statut,
}
2020-09-26 16:19:37 +02:00
if only_primo:
url_kw["only_primo"] = "on"
(
doc,
etuds,
bacs,
bacspecialites,
annee_bacs,
civilites,
statuts,
) = graph_parcours(
2020-09-26 16:19:37 +02:00
formsemestre_id,
only_primo=only_primo,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
)
H = [
2021-06-13 19:12:20 +02:00
html_sco_header.sco_header(
2020-09-26 16:19:37 +02:00
page_title="Parcours étudiants de %(titreannee)s" % sem,
no_side_bar=True,
),
"""<h2 class="formsemestre">Parcours des étudiants de ce semestre</h2>""",
doc,
"<p>%d étudiants sélectionnés</p>" % len(etuds),
2020-09-26 16:19:37 +02:00
_gen_form_selectetuds(
formsemestre_id,
REQUEST=REQUEST,
only_primo=only_primo,
bac=bac,
bacspecialite=bacspecialite,
annee_bac=annee_bac,
civilite=civilite,
2020-09-26 16:19:37 +02:00
statut=statut,
bacs=bacs,
bacspecialites=bacspecialites,
annee_bacs=annee_bacs,
civilites=civilites,
2020-09-26 16:19:37 +02:00
statuts=statuts,
percent=0,
),
"""<p>Origine et devenir des étudiants inscrits dans %(titreannee)s"""
% sem,
"""(<a href="%s">version pdf</a>"""
% url_for("notes.formsemestre_graph_parcours", format="pdf", **url_kw),
""", <a href="%s">image PNG</a>)"""
% url_for("notes.formsemestre_graph_parcours", format="png", **url_kw),
2020-09-26 16:19:37 +02:00
"""</p>""",
"""<p class="help">Le graphe permet de suivre les étudiants inscrits dans le semestre
sélectionné (dessiné en vert). Chaque rectangle représente un semestre (cliquez dedans
pour afficher son tableau de bord). Les flèches indiquent le nombre d'étudiants passant
d'un semestre à l'autre (s'il y en a moins de %s, vous pouvez visualiser leurs noms en
passant la souris sur le chiffre).
</p>"""
% MAX_ETUD_IN_DESCR,
html_sco_header.sco_footer(),
2020-09-26 16:19:37 +02:00
]
return "\n".join(H)
else:
raise ValueError("invalid format: %s" % format)