Skip to content

Commit 6cb5073

Browse files
Revert "1.3.0 (#16)"
This reverts commit 48e3e66.
1 parent 48e3e66 commit 6cb5073

36 files changed

+49381
-4020
lines changed

README.md

Lines changed: 27 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -81,6 +81,28 @@ If you are mainly interested in how getML performs compared to other approaches,
8181
| [SFScores: Predicting health check scores][sfscoresnb] | featuretools | R-squared (getML 29.1%, featuretools 26.5%) |
8282
| [Stats: Predicting users' reputation][statsnb] | featuretools | R-squared (getML 98.1%, featuretools 96.6%) |
8383

84+
### Propositionalization
85+
86+
In particular, we have benchmarked getML's _FastProp_ (short for fast propositionalization) against other implementations of the propositionalization algorithm.
87+
88+
<p align="center" style="text-align: center;">
89+
<img src="propositionalization/comparisons/nrpf_performance.png" />
90+
</p>
91+
92+
As we can see, _FastProp_ is true to its name: It achieves similar or slightly better performance than _featuretools_ or _tsfresh_, but generates features between 11x to 65x faster than these implementations.
93+
94+
If you want to reproduce these results, please refer to the following notebooks:
95+
96+
| | Results | Remarks |
97+
| ------------------------------------ | ------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
98+
| [Air pollution][airpollutionnb_prop] | ~65x faster than featuretools, ~33x faster than tsfresh | The predictive accuracy can be significantly improved by using RelMT instead of propositionalization approaches, please refer to [this notebook][airpollutionnb]. |
99+
| [Dodgers][dodgersnb_prop] | ~42x faster than featuretools, ~75x faster than tsfresh | The predictive accuracy can be significantly improved by using the mapping preprocessor and/or more advanced feature learning algorithms, please refer to [this notebook][dodgersnb]. |
100+
| [Interstate94][interstate94nb_prop] | ~55x faster than featuretools | |
101+
| [Occupancy][occupancynb_prop] | ~87x faster than featuretools, ~41x faster than tsfresh | |
102+
| [Robot][robotnb_prop] | ~162x faster than featuretools, ~77x faster than tsfresh | |
103+
104+
These results are very hardware-dependent and may be different on your machine. However, we have no doubt that you will find that getML's _FastProp_ is significantly faster than _featuretools_ and _tsfresh_ while consuming considerably less memory.
105+
84106
### Relational Dataset Repository
85107

86108
Some benchmarks are also featured on the [Relational Dataset Repository](https://relational.fit.cvut.cz/):
@@ -117,5 +139,10 @@ Some benchmarks are also featured on the [Relational Dataset Repository](https:/
117139
[sfscoresnb]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/sfscores.ipynb
118140
[statsnb]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/stats.ipynb
119141

142+
[airpollutionnb_prop]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/propositionalization/air_pollution_prop.ipynb
143+
[dodgersnb_prop]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/propositionalization/dodgers_prop.ipynb
144+
[interstate94nb_prop]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/propositionalization/interstate94_prop.ipynb
145+
[occupancynb_prop]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/propositionalization/occupancy_prop.ipynb
146+
[robotnb_prop]: https://nbviewer.getml.com/github/getml/getml-demo/blob/master/propositionalization/robot_prop.ipynb
120147

121148

0 commit comments

Comments
 (0)