Time Series Forecasting Challenge
- Tila: Closed
- Palkinto: $100
- Vastaanotetut työt: 112
- Voittaja: DataDKR
Your challenge is to use ML & Deep Learning to build a regression model that can best predict "target" for the "[login to view URL]" data set.
- There are (2) data sets provided
1) "[login to view URL]" contains 58,125 rows along with each row's "target" value. This data set spans from 2019-07-05 09:30:00 to 2020-06-11 11:30:00. This data set should be used to train and validate your model.
2) "[login to view URL]" contains 6,450 rows WITHOUT each row's "target" value. This data set should be used for prediction and for entry submission.
Submission and Performance Criteria:
- Performance against other contestants will be judged based on MEAN ABSOLUTE ERROR of the "[login to view URL]" predictions.
- You will submit a list of predictions in the format "[4.23, 6.61, 26.11, 44.15, ... ]" with one target per row in "[login to view URL]", for a total of 6450 list items. This should be saved as "[login to view URL]" and emailed to contest@[login to view URL] along with your name and screen shot of your MEAN ABSOLUTE ERROR.
- If there is a tie, all top positioned contestants will get the guaranteed payout.
- You may submit as many entries as you wish.
More about the data:
- The data provided is time series data from 15 sensors.
- The column "time_stamp" is the row's time
- Each row represents data from a single sensor, the "sensor_number" column
- There are 15 sensors, thus there are 15 rows per timestamp
- The "target" value is actually "measurement_future_target_1" 4 days in the future
- The sensors are spatially related, so an architecture incorporating CNN or convolutions may be helpful
- The time series data likely has long memory (15 - 60 days), so an architecture incorporating memory or attention (RNN / LSTM / Attention) may be helpful
* As this is time series data it is very important to not shuffle the rows during training as doing so will leak future information into the model *
- You are encouraged to use Python for model construction.
- You may use any regression technique as you see fit (Deep Learning, Machine Learning)
The competition will run for 14 days.
If you win the contest, you may be hired as a consultant once the contest has concluded. Your consultation services will be compensated at an agreed upon payment rate.
“Professional, responsive, and an overall great person.”
wirepulse, United States.