def run_model(model_A, model_B, data_set_name, mode, cutoff_filename=""):
"""
Runs the most recent update model for each day
in the dataset and returns the result to a
csv file in the python directory.
"""
# Load list of folder names which each contain a day
if data_set_name == "InitialTrainingSet_rev1":
days_list = fn.folder_names_init_set()
elif data_set_name == "PublicLeaderboardSet":
days_list = fn.folder_names_test_set()
else:
days_list = []
print "Problem with data set name!"
# Fin_X contains the final predictions for model X
fin_A = fd.FlightPredictions()
# If we have a second model to compare against, to check for
# improvements or anything like that load it into B
if model_B != None:
fin_B = fd.FlightPredictions()
print "Using mode: {}".format(mode)
print "Using data from {}".format(data_set_name)
# Loop through all of the days in the data set
for i, d in enumerate(days_list):
if model_B == None:
print "Running model '{}' on day {} (day {} of {}):".format(model_A, d, i + 1, len(days_list))
else:
print "Running models '{}', '{}' on day {} (day {} of {}):".format(model_A,
model_B, d, i + 1, len(days_list))
# Initialize all the information about each day.
# Extended means we include the flight history events file
# day = efd.ExtendedFlightDay(d, data_set_name, mode, cutoff_filename)
day = fd.FlightDay(d, data_set_name, mode, cutoff_filename)
# day = ad.ASDIDay(d, data_set_name, mode, cutoff_filename)
# day = ead.ExtendedASDIDay(d, data_set_name, mode, cutoff_filename)
# Compute the predicitons for the day
fin_A = return_predictions(model_A, day, fin_A)
if model_B != None:
fin_B = return_predictions(model_B, day, fin_B)
print "\tDay {} has finished".format(d)
print ""
print "All days in {} are done!".format(data_set_name)
if "leaderboard" in mode:
# In leaderboard mode we just write the predictions to a csv file
# for submission to kaggle
fin_A.flight_predictions = fin_A.flight_predictions.sort(columns='flight_history_id')
fin_A.flight_predictions.to_csv('test.csv', index=False)
print "Predictions written to csv file in Python folder."
if model_B != None:
print "Warning: we have disregarded the output of '{}'!".format(model_B)
elif "training" in mode:
# In training mode we can calculate the root mean squared error as we
# know the true values
score_A = rmse.calculate_rmse_score(fin_A.flight_predictions, fin_A.test_data)
if model_B != None:
score_B = rmse.calculate_rmse_score(fin_B.flight_predictions, fin_B.test_data)
else:
score_B = None
scores = {str(model_A): score_A,
str(model_B): score_B}
# Write the scores to the score log for record keeping
# See if we are making improvements
log_predictions(day, model_A, model_B, scores, "scores.log")
return scores
else:
print "Not an option!"
def r_forest():
[X_train, y_train, ind_train] = load_and_format_data('all_combined_no_dates')
[X_test, y_test, ind_test] = load_and_format_data('all_combined_no_dates')
print len(X_train)
# y_train_runway = y_train['actual_runway_arrival_minutes_after_midnight']
# y_train_gate = y_train['actual_gate_arrival_minutes_after_midnight']
forest = RandomForestRegressor(n_estimators=2, random_state=None, n_jobs=-1)
# forest = ExtraTreesRegressor(n_estimators=200, random_state=None, n_jobs=-1)
# forest = GradientBoostingRegressor(n_estimators=200,
# learn_rate=0.1, max_depth=5, random_state=None, loss='ls')
# forest.fit(X_train, y_train_runway)
# y_pred_runway = forest.predict(X_pred)
# forest.fit(X_train, y_train_gate)
# y_pred_gate = forest.predict(X_pred)
# forest.fit(X_train, y_train)
# y_pred = forest.predict(X_pred)
# y_pred_runway = y_pred[:,0]
# y_pred_gate = y_pred[:,1]
# pred = fd.FlightPredictions()
# pred.flight_predictions = pred.flight_predictions.reindex(range(len(ind_pred)))
# pred.flight_predictions['flight_history_id'] = ind_pred
# pred.flight_predictions['actual_runway_arrival'] = y_pred_runway
# pred.flight_predictions['actual_gate_arrival'] = y_pred_gate
# pred.flight_predictions = pred.flight_predictions.sort(columns='flight_history_id')
# pred.flight_predictions.to_csv('test_rand_forest.csv', index=False)
score = []
kfold = cross_validation.KFold(n=len(X_train), k=2, indices=False, shuffle=True)
for i, (traincv, testcv) in enumerate(kfold):
print i
pred = fd.FlightPredictions()
y_pred = []; ind_pred = []; y_pred_runway = []; y_pred_gate = []
print "Starting training...",
forest.fit(X_train[traincv], y_train[traincv])
print "done"
print "Starting prediction...",
y_pred = forest.predict(X_train[testcv])
print "done"
ind_pred = ind_train[testcv].values
y_pred_runway = y_pred[:,0]
y_pred_gate = y_pred[:,1]
pred.flight_predictions = \
pred.flight_predictions.reindex(range(len(ind_pred)))
pred.test_data = \
pred.test_data.reindex(range(len(ind_pred)))
pred.flight_predictions['flight_history_id'] = ind_pred
pred.flight_predictions['actual_runway_arrival'] = y_pred_runway
pred.flight_predictions['actual_gate_arrival'] = y_pred_gate
pred.test_data['flight_history_id'] = ind_pred
pred.test_data['actual_runway_arrival'] = \
y_train['actual_runway_arrival_minutes_after_midnight'][testcv].values
pred.test_data['actual_gate_arrival'] = \
y_train['actual_gate_arrival_minutes_after_midnight'][testcv].values
score.append(rmse.calculate_rmse_score(pred.flight_predictions, pred.test_data))
print score
print np.mean(score)
print np.std(score)
def r_forest():
[X_train, y_train,
ind_train] = load_and_format_data('all_combined_no_dates')
[X_test, y_test, ind_test] = load_and_format_data('all_combined_no_dates')
print len(X_train)
# y_train_runway = y_train['actual_runway_arrival_minutes_after_midnight']
# y_train_gate = y_train['actual_gate_arrival_minutes_after_midnight']
forest = RandomForestRegressor(n_estimators=2,
random_state=None,
n_jobs=-1)
# forest = ExtraTreesRegressor(n_estimators=200, random_state=None, n_jobs=-1)
# forest = GradientBoostingRegressor(n_estimators=200,
# learn_rate=0.1, max_depth=5, random_state=None, loss='ls')
# forest.fit(X_train, y_train_runway)
# y_pred_runway = forest.predict(X_pred)
# forest.fit(X_train, y_train_gate)
# y_pred_gate = forest.predict(X_pred)
# forest.fit(X_train, y_train)
# y_pred = forest.predict(X_pred)
# y_pred_runway = y_pred[:,0]
# y_pred_gate = y_pred[:,1]
# pred = fd.FlightPredictions()
# pred.flight_predictions = pred.flight_predictions.reindex(range(len(ind_pred)))
# pred.flight_predictions['flight_history_id'] = ind_pred
# pred.flight_predictions['actual_runway_arrival'] = y_pred_runway
# pred.flight_predictions['actual_gate_arrival'] = y_pred_gate
# pred.flight_predictions = pred.flight_predictions.sort(columns='flight_history_id')
# pred.flight_predictions.to_csv('test_rand_forest.csv', index=False)
score = []
kfold = cross_validation.KFold(n=len(X_train),
k=2,
indices=False,
shuffle=True)
for i, (traincv, testcv) in enumerate(kfold):
print i
pred = fd.FlightPredictions()
y_pred = []
ind_pred = []
y_pred_runway = []
y_pred_gate = []
print "Starting training...",
forest.fit(X_train[traincv], y_train[traincv])
print "done"
print "Starting prediction...",
y_pred = forest.predict(X_train[testcv])
print "done"
ind_pred = ind_train[testcv].values
y_pred_runway = y_pred[:, 0]
y_pred_gate = y_pred[:, 1]
pred.flight_predictions = \
pred.flight_predictions.reindex(range(len(ind_pred)))
pred.test_data = \
pred.test_data.reindex(range(len(ind_pred)))
pred.flight_predictions['flight_history_id'] = ind_pred
pred.flight_predictions['actual_runway_arrival'] = y_pred_runway
pred.flight_predictions['actual_gate_arrival'] = y_pred_gate
pred.test_data['flight_history_id'] = ind_pred
pred.test_data['actual_runway_arrival'] = \
y_train['actual_runway_arrival_minutes_after_midnight'][testcv].values
pred.test_data['actual_gate_arrival'] = \
y_train['actual_gate_arrival_minutes_after_midnight'][testcv].values
score.append(
rmse.calculate_rmse_score(pred.flight_predictions, pred.test_data))
print score
print np.mean(score)
print np.std(score)