/// <summary>
/// Trains the pipeline with data coming from a <see cref="IDataView"/>.
/// </summary>
public ScikitPipeline Train(IDataView data,
string feature = "Feature", string label = null,
string weight = null, string groupId = null)
{
IDataView trans = data;
using (var ch = _env.Start("Create transforms"))
{
for (int i = 0; i < _transforms.Length; ++i)
{
try
{
trans = _env.CreateTransform(_transforms[i].transformSettings, trans);
}
catch (Exception e)
{
if (e.ToString().Contains("Unknown loadable class"))
{
var nn = _env.ComponentCatalog.GetAllClasses().Length;
var filt = _env.ComponentCatalog.GetAllClasses()
.Select(c => c.UserName)
.OrderBy(c => c)
.Where(c => c.Trim().Length > 2);
var regis = string.Join("\n", filt);
throw Contracts.Except(e, $"Unable to create transform '{_transforms[i].transformSettings}', assembly not registered among {nn}\n{regis}");
}
throw e;
}
_transforms[i].transform = trans as IDataTransform;
}
}
if (_predictor != null)
{
using (var ch = _env.Start("Create Predictor"))
{
_predictor.trainer = TrainerHelper.CreateTrainer(_env, _predictor.trainerSettings);
_roles = new List <KeyValuePair <RoleMappedSchema.ColumnRole, string> >();
_roles.Add(new KeyValuePair <RoleMappedSchema.ColumnRole, string>(RoleMappedSchema.ColumnRole.Feature, feature));
if (!string.IsNullOrEmpty(label))
{
_roles.Add(new KeyValuePair <RoleMappedSchema.ColumnRole, string>(RoleMappedSchema.ColumnRole.Label, label));
}
if (!string.IsNullOrEmpty(groupId))
{
_roles.Add(new KeyValuePair <RoleMappedSchema.ColumnRole, string>(RoleMappedSchema.ColumnRole.Group, groupId));
}
if (!string.IsNullOrEmpty(weight))
{
_roles.Add(new KeyValuePair <RoleMappedSchema.ColumnRole, string>(RoleMappedSchema.ColumnRole.Weight, weight));
}
var roleMap = new RoleMappedData(trans, label, feature, group: groupId, weight: weight);
_predictor.predictor = _predictor.trainer.Train(_env, ch, roleMap);
_predictor.roleMapData = roleMap;
}
}
else
{
_predictor = new StepPredictor()
{
predictor = null,
trainer = null,
trainerSettings = null,
roleMapData = new RoleMappedData(trans)
};
// We predict one to make sure everything works fine.
using (var ch = _env.Start("Compute one prediction."))
{
var df = DataFrameIO.ReadView(trans, 1, keepVectors: true, env: _env);
if (df.Length == 0)
{
throw _env.ExceptEmpty("Something went wrong. The pipeline does not produce any output.");
}
}
}
return(this);
}