data images

README.md

@@ -66,7 +66,21 @@ Using sequence to sequence interfaces for timeseries regression
   </tbody>
 </table>
 
-Project Organization
+## Datasets
+
+To ensure a robust score we use multiple multivariate regression timeseries.
+
+For more see [notebooks/01.0-mc-datasets.ipynb](notebooks/01.0-mc-datasets.ipynb)
+
+![](reports/figures/data_batches_appliances.png)
+![](reports/figures/data_batches_currents.png)
+![](reports/figures/data_batches_gas.png)
+![](reports/figures/data_batches_pm25.png)
+![](reports/figures/data_batches_traffice.png)
+
+
+
+## Project Organization
 ------------
 
     ├── LICENSE
@@ -84,7 +98,7 @@ Project Organization
     ├── reports            <- Generated analysis as HTML, PDF, LaTeX, etc.
     │   └── figures        <- Generated graphics and figures to be used in reporting
     │
-    ├── requirements.txt   <- The requirements file for reproducing the analysis environment, e.g.
+    ├── requirements       <- The requirements folder for reproducing the analysis environment, e.g.
     │                         generated with `pip freeze > requirements.txt`
     │
     ├── setup.py           <- makes project pip installable (pip install -e .) so src can be imported
@@ -96,3 +110,7 @@ Project Organization
 --------
 
 <p><small>Project based on the <a target="_blank" href="https://drivendata.github.io/cookiecutter-data-science/">cookiecutter data science project template</a>. #cookiecutterdatascience</small></p>
+
+```python
+
+```

seq2seq_time/data/data.py

@@ -82,6 +82,9 @@ class RegressionForecastData:
     def __repr__(self):
         return f'<{type(self).__name__} {self.df.shape if (self.df is not None) else None}>'
 
+    def __len__(self):
+        return len(self.df.dropna(subset=self.columns_target))
+
 class GasSensor(RegressionForecastData):
     """
     See: http://archive.ics.uci.edu/ml/datasets/Gas+sensor+array+temperature+modulation
@@ -330,7 +333,6 @@ class IMOSCurrentsVel(RegressionForecastData):
         # made in previous notebook
         xd = xr.load_dataset(outfile)
         df = xd.to_dataframe()
-        df['SPD'] = np.sqrt(df.VCUR**2 + df.UCUR**2)
         df = df[['VCUR',  'UCUR', 'WCUR', 'TEMP',  'DEPTH', 'M2',
        'S2', 'N2', 'K2', 'K1', 'O1', 'P1', 'Q1', 'M4', 'M6', 'S4', 'MK3', 'MM',
        'SSA', 'SA', 'SPD']]
@@ -340,6 +342,6 @@ class IMOSCurrentsVel(RegressionForecastData):
         has_past = df.SPD.isna().rolling(48).sum()<5
         df = df[has_past]
 
-        df = df.resample('10T').first()
+        df = df.resample('30T').mean()
 
         return df

seq2seq_time/data/dataset.py

@@ -31,7 +31,7 @@ class Seq2SeqDataSet(torch.utils.data.Dataset):
         assert df.index.freq is not None, 'should have freq'
         assert_no_objects(df)
 
-        self.freq = df.index.freq
+        self.freq = df.index.freq.freqstr
         self.df = df.dropna(subset=columns_target).ffill()
 
         self.window_past = window_past