public IEnumerable <IRLVItemDisplay> GetMetricSum(Item[] items)
{
var retVal = new List <IRLVItemDisplay>();
double[] metrics = new double[10];
foreach (var item in items)
{
var itemMetrics = PlanogramOptimizer.GetCalculatedWeightedMetricArray(item, Parent.SimulationSettings);
for (int i = 0; i < itemMetrics.Length; i++)
{
metrics[i] += itemMetrics[i];
}
}
for (int i = 0; i < metrics.Length; i++)
{
retVal.Add(new RLVItemDisplay()
{
Name = $"Metric {i + 1}", Value = metrics[i]
});
}
return(retVal);
}
private IEnumerable <Shelf> GetShelfData(RlmLearnedSessionDetails data)
{
var shelves = new List <Shelf>();
IDictionary <string, RlmIODetails> inputDict = data.Inputs.Where(a => a.CycleScore == 1).ToDictionary(a => a.Value, a => a);
IDictionary <long, RlmIODetails> outputDict = data.Outputs.Where(a => a.CycleScore == 1).ToDictionary(a => a.CaseId, a => a);
int numSlots = simSettings.NumShelves * simSettings.NumSlots;
var shelf = new Shelf();
for (int i = 1; i <= numSlots; i++)
{
var input = inputDict[i.ToString()];
var output = outputDict[input.CaseId];
Item itemReference = itemsCache[Convert.ToInt32(output.Value)];
shelf.Add(itemReference, PlanogramOptimizer.GetCalculatedWeightedMetrics(itemReference, simSettings));
if (i % simSettings.NumSlots == 0)
{
shelves.Add(shelf);
shelf = new Shelf();
}
}
return(shelves);
}
public PlanogramOptResults ScorePilot()
{
PlanogramOptResults retVal = new PlanogramOptResults();
var shelves = new List <Shelf>();
double totalMetricScore = 0;
int slotNumber = 0;
Dictionary <int, int> itemDict = new Dictionary <int, int>();
bool hasExceedMax = false;
for (int i = 0; i < simSettings.NumShelves; i++)
{
Shelf shelfInstance = new Shelf();
for (int p = 0; p < simSettings.NumSlots; p++)
{
var inputs = new BasicMLData(1);
inputs[0] = slotNormalizer.Normalize(slotNumber);
IMLData output = network.Compute(inputs);
int index = Convert.ToInt32(itemNormalizer.DeNormalize(output[0]));
Item itemReference = items[index];
if (!hasExceedMax)
{
hasExceedMax = !(CheckDuplicateItem(itemDict, index));
}
// with the item reference we will also have the Attributes which we need to calculate for the metrics
double itemMetricScore = PlanogramOptimizer.GetCalculatedWeightedMetrics(itemReference, simSettings);
// we add the item's metric score to totalMetricScore(which is our Session score in this case)
// notice that we multplied it with the numFacings since that is how many will also show up on the planogram
totalMetricScore += itemMetricScore;
shelfInstance.Add(itemReference, itemMetricScore);
slotNumber++;
}
shelves.Add(shelfInstance);
}
retVal.Shelves = shelves;
retVal.Score = (hasExceedMax) ? 0 : totalMetricScore;
retVal.TimeElapsed = DateTime.Now - simSettings.StartedOn;
return(retVal);
}
/// <summary>
/// Does training or prediction for a set number of sessions
/// </summary>
/// <param name="sessions">Number of session to train/predict for</param>
/// <param name="learn">Lets the RLM know if we are training or predicting</param>
/// <returns>The final output for the training or prediction</returns>
private PlanogramOptResults Optimize(int sessions, bool learn = true, bool enablePlanogramDisplay = false, CancellationToken?cancelToken = null)
{
var output = new PlanogramOptResultsSettings();
// holds the unique SKUs that are already in the planogram. Ensures there are no duplicate SKUs
var hashedSku = new HashSet <int>();
var itemDict = new Dictionary <int, int>();
var shelves = new List <Shelf>();
double totalMetricScore = 0;
for (int i = 0; i < sessions; i++)
{
// reset for next session
shelves.Clear();
hashedSku.Clear();
itemDict.Clear();
ResetCurrentItemIndexes();
totalMetricScore = 0;
totalSessions++;
DateTime startSession = DateTime.Now;
// starts the session. we need to save the session ID as we will pass it to the cycle.Run later on
long sessionId = network.SessionStart();
int numSlotFlattened = 0;
// iterates for how many number of shelves our planogram has
for (int shelf = 1; shelf <= simSettings.NumShelves; shelf++)
{
// this shelf instance will hold the items the RLM will output
var shelfInstance = new Shelf()
{
Number = shelf
};
int itemIndex;
int numFacings;
// iterates for how many number of slots on each shelf
// notice that the slot is incremented depending on how many number of facings was outputed by the RLM
for (int slot = 1; slot <= simSettings.NumSlots; slot += numFacings)
{
itemIndex = -1;
numFacings = -1;
bool isValid = false;
numSlotFlattened++;
do
{
// create the inputs with their corresponding values
var inputs = new List <RlmIOWithValue>();
//inputs.Add(new RlmIOWithValue(network.Inputs.First(a => a.Name == "Shelf"), shelf.ToString()));
//inputs.Add(new RlmIOWithValue(network.Inputs.First(a => a.Name == "Slot"), slot.ToString()));
inputs.Add(new RlmIOWithValue(network.Inputs.First(a => a.Name == "Slot"), numSlotFlattened.ToString()));
var rlmItemOutput = network.Outputs.FirstOrDefault();
var rlmIdeas = new List <RlmIdea>()
{
new RlmOutputLimiter(rlmItemOutput.ID, currentItemIndexes.Count - 1, GetEquivalentIndex)
};
// runs a cycle with the sessionId passed in
var cycle = new RlmCycle();
var rlmOutput = cycle.RunCycle(network, sessionId, inputs, learn, ideas: rlmIdeas);
// get the outputs
// the RLM outputs the item index so we will get the actual reference to the Item later on and
// the second output is the number of facings that will show up on the plangram
itemIndex = Convert.ToInt32(rlmOutput.CycleOutput.Outputs.First(a => a.Name == "Item").Value);
numFacings = 1;//Convert.ToInt32(rlmOutput.CycleOutput.Outputs.First(a => a.Name == "NumFacings").Value);
// we calculate how many remaining slots are there left to check for validity
// because there might not be enough slots for what the number of facings was outputed by the RLM
int remainingSlots = simSettings.NumSlots - shelfInstance.Items.Count();
bool isWithinLimit = CheckDuplicateItem(itemDict, itemIndex, numFacings);
// here we check if the item is not a duplicate and that we have enough slots to fit the number of facings
//if (hashedSku.Add(itemIndex) && remainingSlots >= numFacings)
if (isWithinLimit && remainingSlots >= numFacings)
{
isValid = true;
Item itemReference = items[itemIndex]; // we get the item reference using the index outputed by the RLM
// with the item reference we will also have the Attributes which we need to calculate for the metrics
double itemMetricScore = PlanogramOptimizer.GetCalculatedWeightedMetrics(itemReference, simSettings);
// we add the item's metric score to totalMetricScore(which is our Session score in this case)
// notice that we multplied it with the numFacings since that is how many will also show up on the planogram
totalMetricScore += (itemMetricScore * numFacings);
// add the items to the shelf container depending on how many facings
for (int n = 0; n < numFacings; n++)
{
shelfInstance.Add(itemReference, itemMetricScore);
}
// A non-duplicate is good.
network.ScoreCycle(rlmOutput.CycleOutput.CycleID, 1);
}
else
{
// we give the cycle a zero (0) score as it was not able to satisfy our conditions (punish it)
isValid = false;
network.ScoreCycle(rlmOutput.CycleOutput.CycleID, -1);
//System.Diagnostics.Debug.WriteLine("try again");
}
} while (!isValid); // if invalid, we redo the whole thing until the RLM is able to output an item that is unique and fits the remaining slot in the planogram
}
shelves.Add(shelfInstance);
}
// ends the session with the summed metric score for all items in the planogram
network.SessionEnd(totalMetricScore);
System.Diagnostics.Debug.WriteLine($"Session #{i}, Score: {totalMetricScore}");
// set statistics and the optimized planogram shelves (and items inside them)
output.Shelves = shelves;
metricScoreHistory.Add(totalMetricScore);
metricAvgLastTen.Enqueue(totalMetricScore);
if (metricAvgLastTen.Count > 10)
{
metricAvgLastTen.Dequeue();
}
output.Score = totalMetricScore;
output.AvgScore = metricScoreHistory.Average();
output.AvgLastTen = metricAvgLastTen.Average();
output.MinScore = metricScoreHistory.Min();
output.MaxScore = metricScoreHistory.Max();
output.TimeElapsed = DateTime.Now - simSettings.StartedOn;
output.CurrentSession = totalSessions;
output.MaxItems = MAX_ITEMS;
output.StartRandomness = network.StartRandomness;
output.EndRandomness = network.EndRandomness;
output.SessionsPerBatch = DEFAULT_SESSIONS_PER_BATCH;
output.InputType = inputType.ToString();
output.CurrentRandomnessValue = network.RandomnessCurrentValue;
if (logger != null)
{
SimulationData logdata = new SimulationData();
logdata.Session = output.CurrentSession;
logdata.Score = output.Score;
logdata.Elapse = DateTime.Now - startSession;
logger.Add(logdata);
}
// update the numScoreHits if the sim type is Score
if (simSettings.SimType == SimulationType.Score)
{
if (totalMetricScore >= simSettings.Score.Value)
{
numScoreHits++;
output.NumScoreHits = numScoreHits;
}
else
{
numScoreHits = 0;
}
}
//OnSessionDone?.Invoke(output);
// updates the results to the UI
//bool enableSimDisplay = (!learn) ? true : (learn && simSettings.EnableSimDisplay) ? true : false;
if (enablePlanogramDisplay)
{
output.MetricMin = simSettings.ItemMetricMin;
output.MetricMax = simSettings.ItemMetricMax;
output.CalculateItemColorIntensity();
}
UpdateUI?.Invoke(output, enablePlanogramDisplay);
// checks if we have already by passed the time or score that was set
// if we did, then we stop the training and end it abruptly
if ((simSettings.SimType == SimulationType.Time && simSettings.EndsOn.Value <= DateTime.Now) ||
(simSettings.SimType == SimulationType.Score && numScoreHits >= SimulationSettings.NUM_SCORE_HITS))
{
break;
}
if (cancelToken.HasValue && cancelToken.Value.IsCancellationRequested)
{
return(output);
}
}
return(output);
}
private void runSlmBtn_Click(object sender, RoutedEventArgs e)
{
selectedSlotIndex = -1;
_row = -1;
_col = -1;
itemRow = -1;
itemCol = -1;
SimulationPanel simPanel = new SimulationPanel();
simPanel.SetSimSettings(simSettings);
bool?result = simPanel.ShowDialog();
if (result.HasValue && result.Value == true)
{
// resets grid to default
usePerfColor = false;
FillGrid(planogram, Colors.LightGray);
if (headToHead)
{
FillGrid(planogramTensorflow, Colors.LightGray);
}
// disable control buttons
//statusTxt.Text = statusTxtTensor.Text = "";
statusTxtTensor.Text = "Waiting for RLM to finish running...";
EnableControlButtons(false);
// set simulation settings
simSettings.SimType = simPanel.SimType;
simSettings.Sessions = simPanel.Sessions;
simSettings.Hours = simPanel.Hours;
simSettings.Score = simPanel.Score;
simSettings.EnableSimDisplay = simPanel.EnableSimDisplay;
simSettings.DefaultScorePercentage = simPanel.simScoreSlider.Value;
simSettings.HiddenLayers = simPanel.HiddenLayers;
simSettings.HiddenLayerNeurons = simPanel.HiddenLayerNeurons;
targetScoreTxt.Text = "";
if (simSettings.SimType == SimulationType.Score)
{
targetScoreLbl.Visibility = Visibility.Visible;
targetScoreTxt.Visibility = Visibility.Visible;
targetScoreTxt.Text = simSettings.Score.Value.ToString("n");
targetScoreTxt2.Visibility = Visibility.Visible;
targetScoreTxt2.Text = simSettings.Score.Value.ToString("n");
}
else
{
targetScoreLbl.Visibility = Visibility.Hidden;
targetScoreTxt.Visibility = Visibility.Hidden;
targetScoreTxt2.Visibility = Visibility.Hidden;
}
if (simSettings.SimType == SimulationType.Sessions)
{
sessionPerBatchLbl.Visibility = Visibility.Hidden;
sessionPerBatchTxt.Visibility = Visibility.Hidden;
}
else
{
sessionPerBatchLbl.Visibility = Visibility.Visible;
sessionPerBatchTxt.Visibility = Visibility.Visible;
}
Logger.Clear();
string dbIdentifier = "RLM_planogram_" + Guid.NewGuid().ToString("N");
// instantiate visualizer with this window as its parent reference
visualizer = new RLVOutputVisualizer(this);
core = new RLVCore(dbIdentifier);
// subscribe mainwindow to the comparison event
//visualizer.LearningComparisonDisplayResultsEvent += DisplayLearningComparisonResults;
// open temporary RLV container panel
// todo this must be embeded in this Window instead of the temporary container
if (rlvPanel != null)
{
rlvPanel.Close();
}
rlvPanel = new TempRLVContainerPanel(core, visualizer);
//this.Top = 20;
//tmpPanel.Top = this.Top;
//this.Height = tmpPanel.Height;
//tmpPanel.Left = 10;
//this.Left = tmpPanel.Width + tmpPanel.Left;
//tmpPanel.Visibility = Visibility.Hidden;
Task.Run(() =>
{
// get items from db as well as the min and max metric scores as we need that for the calculation later on
Item[] items;
using (PlanogramContext ctx = new PlanogramContext())
{
MockData mock = new MockData(ctx);
items = itemsCache = mock.GetItemsWithAttr();
simSettings.ItemMetricMin = mock.GetItemMinimumScore(simSettings);
simSettings.ItemMetricMax = mock.GetItemMaximumScore(simSettings);
}
// let's tensorflow (or other listeners) know that it should start training
//OnSimulationStart?.Invoke(items, simSettings); //return;
// initialize and start RLM training
optimizer = new PlanogramOptimizer(items, simSettings, this.UpdateRLMResults, this.UpdateRLMStatus, Logger, dbIdentifier);
//optimizer.OnSessionDone += Optimizer_OnSessionDone;
optimizer.StartOptimization(tokenSource.Token);
});
}
}