# TensorFlow/PyTorch + GPU + Docker

The steps described here was made and tested in **Ubuntu 18.04 x64**, and it's main purpose is make easy to prepare environment from scratch to play with Deep Learning on TensorFlow/PyTorch.

## Steps to prepare environment

1. Update system

    `$> sudo apt-get -y update`

2. Install requirements

    `$> sudo apt install -y python3-pip curl`

3. Install getgist

	`$> pip3 install getgist`
    > If you had any issue with getgist REBOOT your system

4. Install Miniconda

  	`$> getgist rodrigocmoraes install-miniconda.sh`

    `$> bash install-miniconda.sh`

    > Execute the steps below manually when necessary:
    > * Enter
    > * yes
    > * Enter
    > * yes

    > After installation execute command below:

    `$> source ~/.`

5. Install NVidia Driver

	`$> getgist rodrigocmoraes install-nvidia-driver.sh`

	`$> bash install-nvidia-driver.sh`

    > Execute the steps below manually when necessary;
    > * Enter

6. Install Docker/Docker Compose

	`$> getgist rodrigocmoraes install-docker.sh`

	`$> bash install-docker.sh`

    `$> sudo usermod -aG docker $USER`

    `$> sudo reboot`


7. Install NVidia Docker

	`$> getgist rodrigocmoraes install-nvidia-docker.sh`

    `$> bash install-nvidia-docker.sh`

    `$> sudo reboot`


8. Create

	    8.1. TensorFlow - GPU:

	    `$> getgist rodrigocmoraes spec-file-tensorflow-gpu.txt`

	    `$> conda create --name tensorflow-gpu --file spec-file-tensorflow-gpu.txt python=3.6.8`

        `$> conda activate tensorflow-gpu`

	    `$> pip install opencv-python`

        `$> conda deactivate`

		8.2. PyTorch - GPU:

		`$> getgist rodrigocmoraes spec-file-pytorch-gpu.txt`

		`$> conda create --name pytorch-gpu --file spec-file-pytorch-gpu.txt python=3.6.8`

        `$> conda activate pytorch-gpu`

		`$> pip install opencv-python future`

        `$> conda deactivate`

## Test environments:

* TensorFlow - GPU:

    ```python
    from tensorflow.python.client import device_lib
    def get_available_gpus():
        local_device_protos = device_lib.list_local_devices()
        return [x.name for x in local_device_protos if x.device_type == 'GPU']
    print(get_available_gpus())
    ```

    > Expected result:
        `>>> ['/device:GPU:0']`


* PyTorch - GPU:

	```python
    import torch
	id = torch.cura.current_device()
	print(id)
 	print(torch.cuda.get_device_name(id))
 	```

## Most used *conda* commands:

* List conda existing conda environment:

	`$> conda env list`

* Activate conda environment:

    `$> conda activate ENVIRONMENT_NAME`

* Deactivate conda environment:

    `$> conda deactivate`

* Install packages into conda environment:

    `$> conda install --name ENVIRONMENT_NAME PACKAGE[==X.YY.ZZ]`

	or

	`$> conda activate ENVIRONMENT_NAME`

	`$> conda install PACKAGE[==X.YY.ZZ]`

	or from *spec-file.txt*

	`$> conda install --name ENVIRONMENT_NAME PACKAGE[==X.YY.ZZ] --file spec-file.txt`

	or from *requirements.txt*

	`$> while read requirement; do conda install --yes $requirement; done < requirements.txt`

* Export environment specification:

	* From **conda** package manager:

    	`$> conda list --explicit > spec-file-${CONDA_DEFAULT_ENV}.txt`

	* From **pip** package manager:

    `$> pip freeze > requirements-${CONDA_DEFAULT_ENV}.txt`

* Clone environment:

	`$> conda create --name NEW_ENV_NAME --clone ENV_THAT_WILL_BE_CLONNED`