Upload folder using huggingface_hub

README.md

@@ -9,12 +9,13 @@ library_name: pytorch
 
 
 
-# PVNet2
+# PVNet India
 
 ## Model Description
 
 <!-- Provide a longer summary of what this model is/does. -->
-This model class uses satellite data, numericl weather predictions, and recent Grid Service Point( GSP) PV power output to forecast the near-term (~8 hours) PV power output at all GSPs. More information can be found in the model repo [1] and experimental notes in [this google doc](https://docs.google.com/document/d/1fbkfkBzp16WbnCg7RDuRDvgzInA6XQu3xh4NCjV-WDA/edit?usp=sharing).
+This model class uses numerical weather predictions from providers such as ECMWF to forecast the PV power in North West India over the next 48 hours. More information can be found in the model repo [1] and experimental notes [here](https://github.com/openclimatefix/PVNet/tree/main/experiments/india).
+
 
 - **Developed by:** openclimatefix
 - **Model type:** Fusion model
@@ -28,23 +29,19 @@ This model class uses satellite data, numericl weather predictions, and recent G
 
 <!-- This should link to a Data Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
 
-The model is trained on data from 2019-2022 and validated on data from 2022-2023. See experimental notes in the [the google doc](https://docs.google.com/document/d/1fbkfkBzp16WbnCg7RDuRDvgzInA6XQu3xh4NCjV-WDA/edit?usp=sharing) for more details.
+The model is trained on data from 2019-2022 and validated on data from 2022-2023. See experimental notes [here](https://github.com/openclimatefix/PVNet/tree/main/experiments/india)
 
 
 ### Preprocessing
 
-Data is prepared with the `ocf_datapipes.training.pvnet` datapipe [2].
+Data is prepared with the `ocf_datapipes.training.pvnet_site` datapipe [2].
 
 
 ## Results
 
 The training logs for the current model can be found here:
- - [https://wandb.ai/openclimatefix/pvnet2.1/runs/v4xgpar9](https://wandb.ai/openclimatefix/pvnet2.1/runs/v4xgpar9)
-
-
-The training logs for all model runs of PVNet2 can be found [here](https://wandb.ai/openclimatefix/pvnet2.1).
+ - [https://wandb.ai/openclimatefix/india_pv_v2/runs/6e9vqmr4](https://wandb.ai/openclimatefix/india_pv_v2/runs/6e9vqmr4)
 
-Some experimental notes can be found at in [the google doc](https://docs.google.com/document/d/1fbkfkBzp16WbnCg7RDuRDvgzInA6XQu3xh4NCjV-WDA/edit?usp=sharing)
 
 
 ### Hardware

config.json

@@ -1,59 +1,80 @@
 {
     "_target_": "pvnet.models.multimodal.multimodal.Model",
     "output_quantiles": [
-        0.02,
         0.1,
-        0.25,
         0.5,
-        0.75,
-        0.9,
-        0.98
+        0.9
     ],
     "interval_minutes": 15,
     "use_weighted_loss": false,
-    "pv_interval_minutes": 15,
     "nwp_encoders_dict": {
         "ecmwf": {
             "_target_": "pvnet.models.multimodal.encoders.encoders3d.ResConv3DNet2",
             "_partial_": true,
-            "in_channels": 12,
+            "in_channels": 13,
             "out_features": 256,
             "n_res_blocks": 4,
             "hidden_channels": 256,
-            "image_size_pixels": 64
+            "image_size_pixels": 42
+        },
+        "mo_global": {
+            "_target_": "pvnet.models.multimodal.encoders.encoders3d.ResConv3DNet2",
+            "_partial_": true,
+            "in_channels": 10,
+            "out_features": 256,
+            "n_res_blocks": 5,
+            "hidden_channels": 256,
+            "image_size_pixels": 50
+        },
+        "gfs": {
+            "_target_": "pvnet.models.multimodal.encoders.encoders3d.ResConv3DNet2",
+            "_partial_": true,
+            "in_channels": 13,
+            "out_features": 256,
+            "n_res_blocks": 3,
+            "hidden_channels": 256,
+            "image_size_pixels": 10
         }
     },
     "pv_encoder": {
         "_target_": "pvnet.models.multimodal.site_encoders.encoders.SingleAttentionNetwork",
         "_partial_": true,
         "num_sites": 1,
-        "out_features": 40,
+        "out_features": 64,
         "num_heads": 4,
-        "kdim": 40,
-        "id_embed_dim": 20
+        "kdim": 64,
+        "id_embed_dim": 64
     },
     "output_network": {
         "_target_": "pvnet.models.multimodal.linear_networks.networks.ResFCNet2",
         "_partial_": true,
-        "fc_hidden_features": 512,
+        "fc_hidden_features": 128,
         "n_res_blocks": 6,
-        "res_block_layers": 2,
+        "res_block_layers": 4,
         "dropout_frac": 0.0
     },
-    "embedding_dim": 128,
+    "embedding_dim": 16,
     "include_sun": true,
     "include_gsp_yield_history": false,
     "forecast_minutes": 2880,
     "history_minutes": 60,
-    "min_sat_delay_minutes": 60,
-    "sat_history_minutes": 90,
     "nwp_history_minutes": {
-        "ecmwf": 60
+        "ecmwf": 60,
+        "mo_global": 60,
+        "gfs": 0
     },
     "nwp_forecast_minutes": {
-        "ecmwf": 2880
+        "ecmwf": 2880,
+        "mo_global": 1680,
+        "gfs": 2220
+    },
+    "nwp_interval_minutes": {
+        "ecmwf": 60,
+        "mo_global": 60,
+        "gfs": 180
     },
     "pv_history_minutes": 60,
+    "pv_interval_minutes": 15,
     "target_key": "pv",
     "optimizer": {
         "_target_": "pvnet.optimizers.EmbAdamWReduceLROnPlateau",

data_config.yaml

@@ -6,8 +6,18 @@ input_data:
   default_history_minutes: 60
   nwp:
     ecmwf:
-      forecast_minutes: 2880
+      coarsen_to_degrees: 0.2
+      dropout_fraction: 1.0
+      dropout_timedeltas_minutes:
+      - -360
+      forecast_minutes: 2880.0
       history_minutes: 60
+      max_staleness_minutes: null
+      nwp_accum_channels:
+      - dswrf
+      - dlwrf
+      - sr
+      - duvrs
       nwp_channels:
       - hcc
       - lcc
@@ -21,18 +31,73 @@ input_data:
       - v10
       - dlwrf
       - dswrf
-      nwp_image_size_pixels_height: 64
-      nwp_image_size_pixels_width: 64
+      - duvrs
+      nwp_image_size_pixels_height: 42
+      nwp_image_size_pixels_width: 42
       nwp_provider: ecmwf
       nwp_zarr_path: PLACEHOLDER.zarr
       time_resolution_minutes: 60
-      x_dim_name: longitude
-      y_dim_name: latitude
+    gfs:
+      dropout_fraction: 1.0
+      dropout_timedeltas_minutes:
+      - -300
+      forecast_minutes: 2160.0
+      history_minutes: 0
+      nwp_channels:
+      - t
+      - prate
+      - u10
+      - v10
+      - dlwrf
+      - dswrf
+      - hcc
+      - lcc
+      - mcc
+      - r
+      - sde
+      - tcc
+      - vis
+      nwp_image_size_pixels_height: 10
+      nwp_image_size_pixels_width: 10
+      nwp_provider: gfs
+      nwp_zarr_path: PLACEHOLDER.zarr
+      time_resolution_minutes: 180
+    mo_global:
+      dropout_fraction: 1.0
+      dropout_timedeltas_minutes:
+      - -180
+      forecast_minutes: 1680.0
+      history_minutes: 60
+      max_staleness_minutes: null
+      nwp_channels:
+      - temperature_sl
+      - wind_u_component_10m
+      - wind_v_component_10m
+      - downward_shortwave_radiation_flux_gl
+      - cloud_cover_high
+      - cloud_cover_low
+      - cloud_cover_medium
+      - relative_humidity_sl
+      - snow_depth_gl
+      - visibility_sl
+      nwp_image_size_pixels_height: 50
+      nwp_image_size_pixels_width: 50
+      nwp_provider: mo_global
+      nwp_zarr_path: PLACEHOLDER.zarr
+      time_resolution_minutes: 60
   pv:
+    dropout_fraction: 1.0
+    dropout_timedeltas_minutes:
+    - -15
+    forecast_minutes: 2880
+    history_minutes: 60
     n_pv_systems_per_example: 1
     pv_files_groups:
     - label: india
       pv_filename: PLACEHOLDER.netcdf
       pv_metadata_filename: PLACEHOLDER.csv
     pv_ml_ids: []
+    system_dropout_fraction_max: 0
+    system_dropout_fraction_min: 0
+    system_dropout_timedeltas_minutes: null
     time_resolution_minutes: 15

pytorch_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1be652c35d18ac4ec55d79363e3f98c145bc81838dcd529984d92af5d2a0ac20
-size 283446466
+oid sha256:70156641924ac90d26bab7157c12cfad57b33a43d092d5e5068d9ec1982d7b82
+size 237540402