Predict the total ride duration of taxi trips in New York City. primary dataset is one released by the NYC Taxi and Limousine Commission, which includes pickup time, geo-coordinates, number of passengers, and several other variables.
Please download the train data via : https://www.kaggle.com/c/nyc-taxi-trip-duration/data, and save at data/train.csv. Then
you should be able to run the ML demo code (scripts under run/)
- Kaggle Page
- Analysis nb - EDA ipython notebook
- ML nb - ML ipython notebook
- Main code - Final ML code in python
- Team member repo
Please also check NYC_Taxi_Pipeline in case you are interested in the data engineering projects with similar taxi dataset.
├── README.md
├── data : directory of train/test data
├── documents: main reference
├── model : save the tuned model
├── notebook : main analysis
├── output : prediction outcome
├── reference: other reference
├── run : fire the train/predict process
├── script : utility script for data preprocess / feature extract / train /predict
├── spark_ : Re-run the modeling with SPARK Mlib framework : JAVA / PYTHON / SCALA
└── start.sh : launch training env
# demo of submit_xgb_387.py
cd NYC_Taxi_Trip_Duration
export PYTHONPATH=/Users/$USER/NYC_Taxi_Trip_Duration/
python run/submit_xgb_core_377_OSRM.py
....
KFold(n_splits=10, random_state=None, shuffle=False)
TRAIN: [ 134528 134529 134530 ... 1345276 1345277 1345278] TEST: [ 0 1 2 ... 134525 134526 134527]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [134528 134529 134530 ... 269053 269054 269055]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [269056 269057 269058 ... 403581 403582 403583]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [403584 403585 403586 ... 538109 538110 538111]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [538112 538113 538114 ... 672637 672638 672639]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [672640 672641 672642 ... 807165 807166 807167]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [807168 807169 807170 ... 941693 941694 941695]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [ 941696 941697 941698 ... 1076221 1076222 1076223]
TRAIN: [ 0 1 2 ... 1345276 1345277 1345278] TEST: [1076224 1076225 1076226 ... 1210749 1210750 1210751]
TRAIN: [ 0 1 2 ... 1210749 1210750 1210751] TEST: [1210752 1210753 1210754 ... 1345276 1345277 1345278]
[0] train-rmse:5.71905 valid-rmse:5.71921
# Multiple eval metrics have been passed: 'valid-rmse' will be used # for early stopping.
# Will train until valid-rmse hasn't improved in 250 rounds.
[15] train-rmse:3.12115 valid-rmse:3.12236
[30] train-rmse:1.7254 valid-rmse:1.72854
[45] train-rmse:0.988145 valid-rmse:0.995816
[60] train-rmse:0.61477 valid-rmse:0.63196
[75] train-rmse:0.440141 valid-rmse:0.470721
[90] train-rmse:0.365692 valid-rmse:0.408737
[105] train-rmse:0.334895 valid-rmse:0.387246
[120] train-rmse:0.320705 valid-rmse:0.379742
[135] train-rmse:0.312278 valid-rmse:0.376631
[150] train-rmse:0.306056 valid-rmse:0.375053
[165] train-rmse:0.300808 valid-rmse:0.374119
[180] train-rmse:0.296447 valid-rmse:0.373308
[195] train-rmse:0.293421 valid-rmse:0.372817
[210] train-rmse:0.290248 valid-rmse:0.372474
[225] train-rmse:0.287413 valid-rmse:0.372054
[240] train-rmse:0.285092 valid-rmse:0.37174
[255] train-rmse:0.282846 valid-rmse:0.37154
[270] train-rmse:0.28117 valid-rmse:0.371426
[285] train-rmse:0.278903 valid-rmse:0.371133
[300] train-rmse:0.276861 valid-rmse:0.370875
[315] train-rmse:0.275129 valid-rmse:0.370712
[330] train-rmse:0.273314 valid-rmse:0.370525
[345] train-rmse:0.271538 valid-rmse:0.370385
[360] train-rmse:0.270056 valid-rmse:0.370277
[375] train-rmse:0.268084 valid-rmse:0.370096
[390] train-rmse:0.266806 valid-rmse:0.370005
[405] train-rmse:0.26524 valid-rmse:0.36985
[420] train-rmse:0.26388 valid-rmse:0.369732
[435] train-rmse:0.262156 valid-rmse:0.369577
[450] train-rmse:0.260847 valid-rmse:0.369475
[465] train-rmse:0.259224 valid-rmse:0.369379
[480] train-rmse:0.257622 valid-rmse:0.369215
[495] train-rmse:0.256402 valid-rmse:0.369111
[499] train-rmse:0.256006 valid-rmse:0.369087
trip_duration
id
id3004672 858.211670
id3505355 591.309570
id1217141 379.027527
id2150126 884.275757
id1598245 378.042847
id0668992 920.350037
id1765014 1413.732666
id0898117 968.557373
id3905224 3472.009277
id1543102 476.564056
id3024712 1047.087036
id3665810 330.427673
id1836461 452.128021
id3457080 630.039307
id3376065 1194.740723
id3008739 828.353821
id0902216 1145.874268
id3564824 519.362488
....
[625134 rows x 1 columns]
# demo of submit_tpot.py
cd NYC_Taxi_Trip_Duration
export PYTHONPATH=/Users/$USER/NYC_Taxi_Trip_Duration/
python run/submit_tpot.py
# output
dropoff_datetime dropoff_latitude dropoff_longitude id \
0 2016-03-14 17:32:30 40.765602 -73.964630 id2875421
1 2016-06-12 00:54:38 40.731152 -73.999481 id2377394
2 2016-01-19 12:10:48 40.710087 -74.005333 id3858529
3 2016-01-30 22:09:03 40.749184 -73.992081 id0801584
4 2016-06-17 22:40:40 40.765896 -73.957405 id1813257
Best pipeline: ExtraTreesRegressor(input_matrix, bootstrap=True, max_features=0.8, min_samples_leaf=4, min_samples_split=8, n_estimators=100)
The cross-validation MSE
-0.1800694337127113
Imputing missing values in feature set
....
{'sklearn.linear_model.ElasticNetCV': {'tol': [1e-05, 0.0001, 0.001, 0.01, 0.1], 'l1_ratio': array([0. , 0.05, 0.1 , 0.15, 0.2 , 0.25, 0.3 , 0.35, 0.4 , 0.45, 0.5 ,
0.55, 0.6 , 0.65, 0.7 , 0.75, 0.8 , 0.85, 0.9 , 0.95, 1. ])}, 'sklearn.cluster.FeatureAgglomeration': {'affinity': ['euclidean', 'l1', 'l2', 'manhattan', 'cosine'], 'linkage': ['ward', 'complete', 'average']}, 'sklearn.preprocessing.Binarizer': {'threshold': array([0. , 0.05, 0.1 , 0.15, 0.2 , 0.25, 0.3 , 0.35, 0.4 , 0.45, 0.5 ,
0.55, 0.6 , 0.65, 0.7 , 0.75, 0.8 , 0.85, 0.9 , 0.95, 1. ])}
.....
EDA -> Data Preprocess -> Model select -> Feature engineering -> Model tune -> Prediction ensemble
# PROJECT WORKFLOW
#### 1. DATA EXPLORATION (EDA)
Analysis : /notebook
#### 2. FEATURE EXTRACTION
2-1. **Feature dependency**
2-2. **Encode & Standardization**
2-3. **Feature transformation**
2-4. **Dimension reduction** ( via PCA)
Script : /script
Modeling : /run
#### 3. PRE-TEST
3-1. **Input all standardized features to all models** <br>
3-2. **Regression**
#### 4. OPTIMIZATION
4-1. **Feature optimization**
4-2. **Super-parameters tuning**
4-3. **Aggregation**
#### 5. RESULTS
-> check the output csv, log
# unit test
$ export PYTHONPATH=/Users/$USER/NYC_Taxi_Trip_Duration/
$ pytest -v tests/
# ============================== test session starts ==============================
# platform darwin -- Python 3.6.4, pytest-5.0.1, py-1.5.2, pluggy-0.12.0 -- /Users/jerryliu/anaconda3/envs/yen_dev/bin/python
# cachedir: .pytest_cache
# rootdir: /Users/jerryliu/NYC_Taxi_Trip_Duration
# plugins: cov-2.7.1
# collected 5 items
# tests/test_data_exist.py::test_training_data_exist PASSED [ 20%]
# tests/test_data_exist.py::test_validate_data_exist PASSED [ 40%]
# tests/test_udf.py::test_get_haversine_distance PASSED [ 60%]
# tests/test_udf.py::test_get_manhattan_distance PASSED [ 80%]
# tests/test_udf.py::test_get_direction PASSED [100%]
# =========================== 5 passed in 2.68 seconds ===========================-
XGBoost
-
LightGBM