Installation¶

Prerequisites¶

Platform¶

This installation instruction has been tested on Linux (Ubuntu Linux 18.04 and 19.04) and macOS (macOS 10.14 Mojave).

Python¶

sccloud works with Python 3 (Python 3.5 or greater) only.

For Ubuntu, install Python 3 by:

sudo apt install python

For macOS, Python 3 can be installed by either Miniconda package manger (see below), Homebrew, or the Python website.

Others¶

For Ubuntu, install the following packages:

sudo apt install build-essential cmake libxml2-dev wget

For macOS, install by:

brew install cmake libxml2 curl

Install via PyPI¶

Linux¶

Assuming Python 3 is installed on your OS. Then install pip for Python 3, along with some extra libraries for building sccloud:

sudo apt install python3-pip

Now install sccloud via pip:

pip3 install sccloud

Alternatively, if you want to use mkl package for speed improvement, type:

pip3 install sccloud[mkl]

If you want to use sccloud’s FIt-SNE feature. First, install fftw library:

sudo apt install libfftw3-dev

Then type:

pip3 install sccloud[fitsne]

Or if you want both extra features, type:

pip3 install sccloud[mkl,fitsne]

macOS¶

pip3 is already installed. Follow the same steps as above for Linux, except that when installing fftw, you have to go to its website for instruction.

Install via Miniconda¶

You can also choose to install sccloud in a dedicated Conda environment without affecting your OS.

Package Manager¶

Linux¶

Use the following commands to install a Miniconda on your system:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh .
export CONDA_PATH=/home/foo
bash Miniconda3-latest-Linux-x86_64.sh -p $CONDA_PATH/miniconda3
mv Miniconda3-latest-Linux-x86_64.sh $CONDA_PATH/miniconda3
source ~/.bashrc

Feel free to change /home/foo to your own directory on handling Miniconda.

macOS¶

Use the following commands:

curl -O https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
export CONDA_PATH=/Users/foo
bash Miniconda3-latest-MacOSX-x86_64.sh -p $CONDA_PATH/miniconda3
mv Miniconda3-latest-MacOSX-x86_64.sh $CONDA_PATH/miniconda3

Feel free to change /Users/foo to your own directory on handling Miniconda.

Install sccloud¶

Both Linux and macOS share this installation step.

Create a conda environment for sccloud. Let sccloud be its name, but you are free to choose your own:
```
conda create -n sccloud -y pip
```

For macOS, you may need to do the following extra commands after creation:

mkdir -p $CONDA_PATH/miniconda3/envs/sccloud/etc/conda/activate.d
mkdir -p $CONDA_PATH/miniconda3/envs/sccloud/etc/conda/deactivate.d
printf '#!/bin/sh\n\nexport KMP_DUPLICATE_LIB_OK=true\n' > $CONDA_PATH/miniconda3/envs/sccloud/etc/conda/activate.d/env_vars.sh
printf '#!/bin/sh\n\nunset KMP_DUPLICATE_LIB_OK' > $CONDA_PATH/miniconda3/envs/sccloud/etc/conda/deactivate.d/env_vars.sh

where $CONDA_PATH is set in the previous step.

Enter conda environment by activating:
```
conda activate sccloud
```

or:

source activate sccloud

(Optional) If you want to use the Intel mkl package for speed improvement, type:
```
conda install -y -c anaconda numpy
```

Also, if you want to use sccloud’s FIt-SNE feature, which depends on fftw package, type:

conda install -y -c conda-forge fftw

Install sccloud:
```
pip install sccloud
```

If you want to use sccloud’s FIt-SNE feature, type:

pip install sccloud[fitsne]

Use sccloud in UGER¶

First, you need to request a RedHat7 server:

qrsh -q interactive -l h_vmem=4g -l os=RedHat7 -P regevlab

Then, if you have installed sccloud, you could activate the virtual environment:

source activate sccloud

Or, you can use an installed version by typing:

source /ahg/regevdata/users/libo/miniconda3/bin/activate sccloud