You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: Tools/PyTorch/TimeSeriesPredictionPlatform/README.md
+12-8Lines changed: 12 additions & 8 deletions
Original file line number
Diff line number
Diff line change
@@ -95,41 +95,45 @@ For those unable to set up the required environment or create your own container
95
95
## Quick start guide
96
96
97
97
### Getting Started
98
-
1. Create a dataset directory. The directory can be arbitrary, and it is recommended not to include it in the TimeSeriesPredictionPlatform directory. This arbitrary directory will be mounted to the TSPP container later. In the following steps this directory will be referred to as /your/datasets/.
98
+
1. Clone the NVIDIA Deep Learning Examples repository:
2. Create a dataset directory. The directory can be arbitrary, and it is recommended not to include it in the TimeSeriesPredictionPlatform directory. This arbitrary directory will be mounted to the TSPP container later. In the following steps this directory will be referred to as /your/datasets/.
99
103
100
-
2. Enter the Deep Learning Examples TSPP repository:
104
+
3. Enter the Deep Learning Examples TSPP repository:
101
105
102
106
```
103
107
cd DeeplearningExamples/Tools/PyTorch/TimeSeriesPredictionPlatform
104
108
```
105
-
3. Run repository setup
109
+
4. Run repository setup
106
110
```
107
111
source scripts/setup.sh
108
112
```
109
113
110
-
3. Build the docker image:
114
+
5. Build the docker image:
111
115
```
112
116
docker build -t tspp .
113
117
```
114
118
115
-
4. Next we will start our container and mount the dataset directory, which means that /workspace/datasets/ points to /your/datasets/. Any changes made to this folder in the docker container are reflected in the original directory and vice versa. If we want to mount additional folders we can add ‘-v /path/on/local/:/path/in/container/’ to the run command. This will be useful if we want to save the outputs from training or inference once we close the container. To start the docker container:
119
+
6. Next we will start our container and mount the dataset directory, which means that /workspace/datasets/ points to /your/datasets/. Any changes made to this folder in the docker container are reflected in the original directory and vice versa. If we want to mount additional folders we can add ‘-v /path/on/local/:/path/in/container/’ to the run command. This will be useful if we want to save the outputs from training or inference once we close the container. To start the docker container:
116
120
```
117
121
docker run -it --gpus all --ipc=host --network=host -v /your/datasets/:/workspace/datasets/ tspp bash
118
122
```
119
123
120
-
5. After running the previous command you will be placed inside the docker container in the /workspace directory. Inside the container, download either the electricity or traffic dataset:
124
+
7. After running the previous command you will be placed inside the docker container in the /workspace directory. Inside the container, download either the electricity or traffic dataset:
The raw electricity dataset is the 15 minute electricity consumption of 370 customers from the UCI Electricity Load Diagrams. We aggregate to an hourly forecast and use the previous week to predict the following day.
125
129
The raw traffic dataset is the 10 minute occupancy rate of San Francisco freeways from 440 sensors downloaded from the UCI PEMS-SF Data Set. We again aggregate to an hourly forecast and use the previous week to predict the following day.
126
130
127
-
6. Preprocess the dataset:
131
+
8. Preprocess the dataset:
128
132
```
129
133
python launch_preproc.py dataset={dataset}
130
134
```
131
135
132
-
7. Launch the training, validation, and testing process using the temporal fusion transformer model:
136
+
9. Launch the training, validation, and testing process using the temporal fusion transformer model:
0 commit comments