From 8013cf62a02aff9f57c892a3fd9f0d83bc80a6ce Mon Sep 17 00:00:00 2001 From: Lewis Woolfson Date: Sun, 31 May 2020 14:38:54 +0100 Subject: [PATCH] initial model formulation --- modelling/scheduler.py | 263 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 263 insertions(+) create mode 100644 modelling/scheduler.py diff --git a/modelling/scheduler.py b/modelling/scheduler.py new file mode 100644 index 0000000..bf8474c --- /dev/null +++ b/modelling/scheduler.py @@ -0,0 +1,263 @@ +import pandas as pd +import pyomo.environ as pe +import pyomo.gdp as pyogdp +import os +import matplotlib.pyplot as plt +import matplotlib.cm as cm +from itertools import product + + +# check with Taha if code is too similar to Alstom? +class TheatreScheduler: + + def __init__(self, case_file_path, session_file_path): + """ + Read case and session data into Pandas DataFrames + Args: + case_file_path (str): path to case data in CSV format + session_file_path (str): path to theatre session data in CSV format + """ + try: + self.df_cases = pd.read_csv(case_file_path) + except FileNotFoundError: + print("Case data not found.") + + try: + self.df_sessions = pd.read_csv(session_file_path) + except FileNotFoundError: + print("Session data not found") + + self.model = self.create_model() + + def _generate_case_durations(self): + """ + Generate mapping of cases IDs to median case time for the procedure + Returns: + (dict): dictionary with CaseID as key and median case time (mins) for procedure as value + """ + return pd.Series(self.df_cases["Median Duration"].values, index=self.df_cases["CaseID"]).to_dict() + + def _generate_session_durations(self): + """ + Generate mapping of all theatre sessions IDs to session duration in minutes + Returns: + (dict): dictionary with SessionID as key and session duration as value + """ + return pd.Series(self.df_sessions["Duration"].values, index=self.df_sessions["SessionID"]).to_dict() + + def _generate_session_start_times(self): + """ + Generate mapping from SessionID to session start time + Returns: + (dict): dictionary with SessionID as key and start time in minutes since midnight as value + """ + # Convert session start time from HH:MM:SS format into seconds elapsed since midnight + self.df_sessions.loc[:, "Start"] = pd.to_timedelta(self.df_sessions["Start"]) + self.df_sessions.loc[:, "Start"] = self.df_sessions["Start"].dt.total_seconds() / 60 + return pd.Series(self.df_sessions["Start"].values, index=self.df_sessions["SessionID"]).to_dict() + + def _get_ordinal_case_deadlines(self): + """ + #TODO + Returns: + + """ + self.df_cases.loc[:, "TargetDeadline"] = pd.to_datetime(self.df_cases["TargetDeadline"], format="%d/%m/%Y") + self.df_cases.loc[:, "TargetDeadline"] = self.df_cases["TargetDeadline"].apply(lambda date: date.toordinal()) + return pd.Series(self.df_cases["TargetDeadline"].values, index=self.df_cases["CaseID"]).to_dict() + + def _get_ordinal_session_dates(self): + """ + #TODO + Returns: + + """ + self.df_sessions.loc[:, "Date"] = pd.to_datetime(self.df_sessions["Date"], format="%d/%m/%Y") + self.df_sessions.loc[:, "Date"] = self.df_sessions["Date"].apply(lambda date: date.toordinal()) + return pd.Series(self.df_sessions["Date"].values, index=self.df_sessions["SessionID"]).to_dict() + + def _generate_disjunctions(self): + """ + #TODO + Returns: + disjunctions (list): list of tuples containing disjunctions + """ + cases = self.df_cases["CaseID"].to_list() + sessions = self.df_sessions["SessionID"].to_list() + disjunctions = [] + for (case1, case2, session) in product(cases, cases, sessions): + if (case1 != case2) and (case2, case1, session) not in disjunctions: + disjunctions.append((case1, case2, session)) + + return disjunctions + + def create_model(self): + model = pe.ConcreteModel() + + # Model Data + + # List of case IDs in surgical waiting list + model.CASES = pe.Set(initialize=self.df_cases["CaseID"].tolist()) + # List of sessions IDs + model.SESSIONS = pe.Set(initialize=self.df_sessions["SessionID"].tolist()) + # List of tasks - all possible (caseID, sessionID) combination + model.TASKS = pe.Set(initialize=model.CASES * model.SESSIONS, dimen=2) + # The duration (median case time) for each operation + model.CASE_DURATION = pe.Param(model.CASES, initialize=self._generate_case_durations()) + # The duration of each theatre session + model.SESSION_DURATION = pe.Param(model.SESSIONS, initialize=self._generate_session_durations()) + # The start time of each theatre session + model.SESSION_START_TIME = pe.Param(model.SESSIONS, initialize=self._generate_session_start_times()) + # The deadline of each case + model.CASE_DEADLINES = pe.Param(model.CASES, initialize=self._get_ordinal_case_deadlines()) + # The date of each theatre session + model.SESSION_DATES = pe.Param(model.SESSIONS, initialize=self._get_ordinal_session_dates()) + + + model.DISJUNCTIONS = pe.Set(initialize=self._generate_disjunctions(), dimen=3) + + ub = 1440 # seconds in a day + model.M = pe.Param(initialize=1e3*ub) # big M + max_util = 0.85 + num_cases = self.df_cases.shape[0] + + # Decision Variables + model.SESSION_ASSIGNED = pe.Var(model.TASKS, domain=pe.Binary) + model.CASE_START_TIME = pe.Var(model.TASKS, bounds=(0, ub), within=pe.PositiveReals) + model.CASES_IN_SESSION = pe.Var(model.SESSIONS, bounds=(0, num_cases), within=pe.PositiveReals) + + # Objective + def objective_function(model): + return pe.summation(model.CASES_IN_SESSION) + #return sum([model.SESSION_ASSIGNED[case, session] for case in model.CASES for session in model.SESSIONS]) + model.OBJECTIVE = pe.Objective(rule=objective_function, sense=pe.maximize) + + # Constraints + + # Case start time must be after start time of assigned theatre session + def case_start_time(model, case, session): + return model.CASE_START_TIME[case, session] >= model.SESSION_START_TIME[session] - \ + ((1 - model.SESSION_ASSIGNED[(case, session)])*model.M) + model.CASE_START = pe.Constraint(model.TASKS, rule=case_start_time) + + # Case end time must be before end time of assigned theatre session + def case_end_time(model, case, session): + return model.CASE_START_TIME[case, session] + model.CASE_DURATION[case] <= model.SESSION_START_TIME[session] + \ + model.SESSION_DURATION[session]*max_util + ((1 - model.SESSION_ASSIGNED[(case, session)]) * model.M) + model.CASE_END_TIME = pe.Constraint(model.TASKS, rule=case_end_time) + + # Cases can be assigned to a maximum of one session + def session_assignment(model, case): + return sum([model.SESSION_ASSIGNED[(case, session)] for session in model.SESSIONS]) <= 1 + model.SESSION_ASSIGNMENT = pe.Constraint(model.CASES, rule=session_assignment) + + def set_deadline_condition(model, case, session): + return model.SESSION_DATES[session] <= model.CASE_DEADLINES[case] + ((1 - model.SESSION_ASSIGNED[case, session])*model.M) + model.APPLY_DEADLINE = pe.Constraint(model.TASKS, rule=set_deadline_condition) + + def no_case_overlap(model, case1, case2, session): + return [model.CASE_START_TIME[case1, session] + model.CASE_DURATION[case1] <= model.CASE_START_TIME[case2, session] + \ + ((2 - model.SESSION_ASSIGNED[case1, session] - model.SESSION_ASSIGNED[case2, session])*model.M), + model.CASE_START_TIME[case2, session] + model.CASE_DURATION[case2] <= model.CASE_START_TIME[case1, session] + \ + ((2 - model.SESSION_ASSIGNED[case1, session] - model.SESSION_ASSIGNED[case2, session])*model.M)] + + model.DISJUNCTIONS_RULE = pyogdp.Disjunction(model.DISJUNCTIONS, rule=no_case_overlap) + + def theatre_util(model, session): + return model.CASES_IN_SESSION[session] == \ + sum([model.SESSION_ASSIGNED[case, session] for case in model.CASES]) + + model.THEATRE_UTIL = pe.Constraint(model.SESSIONS, rule=theatre_util) + + pe.TransformationFactory("gdp.bigm").apply_to(model) + + return model + + def solve(self, solver_name, options=None, solver_path=None, local=True): + + if solver_path is not None: + solver = pe.SolverFactory(solver_name, executable=solver_path) + else: + solver = pe.SolverFactory(solver_name) + + # TODO remove - too similar to alstom + if options is not None: + for key, value in options.items(): + solver.options[key] = value + + if local: + solver_results = solver.solve(self.model, tee=True) + else: + solver_manager = pe.SolverManagerFactory("neos") + solver_results = solver_manager.solve(self.model, opt=solver) + + results = [{"Case": case, + "Session": session, + "Session Date": self.model.SESSION_DATES[session], + "Case Deadline": self.model.CASE_DEADLINES[case], + "Days before deadline": self.model.CASE_DEADLINES[case] - self.model.SESSION_DATES[session], + "Start": self.model.CASE_START_TIME[case, session](), + "Assignment": self.model.SESSION_ASSIGNED[case, session]()} + for (case, session) in self.model.TASKS] + + self.df_times = pd.DataFrame(results) + + all_cases = self.model.CASES.value_list + cases_assigned = [] + for (case, session) in self.model.SESSION_ASSIGNED: + if self.model.SESSION_ASSIGNED[case, session] == 1: + cases_assigned.append(case) + + cases_missed = list(set(all_cases).difference(cases_assigned)) + print("Number of cases assigned = {} out of {}:".format(len(cases_assigned), len(all_cases))) + print("Cases assigned: ", cases_assigned) + print("Number of cases missed = {} out of {}:".format(len(cases_missed), len(all_cases))) + print("Cases missed: ", cases_missed) + self.model.CASES_IN_SESSION.pprint() + print("Total Objective = {}".format(sum(self.model.CASES_IN_SESSION.get_values().values()))) + print("Number of constraints = {}".format(solver_results["Problem"].__getitem__(0)["Number of constraints"])) + #self.model.SESSION_ASSIGNED.pprint() + print(self.df_times[self.df_times["Assignment"] == 1].to_string()) + self.draw_gantt() + + def draw_gantt(self): + + df = self.df_times[self.df_times["Assignment"] == 1] + cases = sorted(list(df['Case'].unique())) + sessions = sorted(list(df['Session'].unique())) + + bar_style = {'alpha': 1.0, 'lw': 25, 'solid_capstyle': 'butt'} + text_style = {'color': 'white', 'weight': 'bold', 'ha': 'center', 'va': 'center'} + colors = cm.Dark2.colors + + df.sort_values(by=['Case', 'Session']) + df.set_index(['Case', 'Session'], inplace=True) + + fig, ax = plt.subplots(1, 1) + for c_ix, c in enumerate(cases, 1): + for s_ix, s in enumerate(sessions, 1): + if (c, s) in df.index: + xs = df.loc[(c, s), 'Start'] + xf = df.loc[(c, s), 'Start'] + \ + self.df_cases[self.df_cases["CaseID"] == c]["Median Duration"] + ax.plot([xs, xf], [s] * 2, c=colors[c_ix % 7], **bar_style) + ax.text((xs + xf) / 2, s, c, **text_style) + + ax.set_title('Assigning Ophthalmology Cases to Theatre Sessions') + ax.set_xlabel('Time') + ax.set_ylabel('Sessions') + ax.grid(True) + + fig.tight_layout() + plt.show() + + +if __name__ == "__main__": + case_path = os.path.join(os.path.dirname(os.getcwd()), "data", "cases.csv") + session_path = os.path.join(os.path.dirname(os.getcwd()), "data", "sessions.csv") + cbc_path = "C:\\Users\\LONLW15\\Documents\\Linear Programming\\Solvers\\cbc.exe" + + options = {"seconds": 300} + scheduler = TheatreScheduler(case_file_path=case_path, session_file_path=session_path) + scheduler.solve(solver_name="cbc", solver_path=cbc_path, options=options) + #scheduler.solve(solver_name="cbc", local=False, options=None)