> ## Documentation Index
> Fetch the complete documentation index at: https://felimet-hub.jmcores.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Conda Installation and Configuration Guide
> Conda guide covering Anaconda vs. Miniconda, installation, custom paths, channels, and Python environment management for data science workflows.
`Windows 10/11` `Miniconda` `Anaconda`
Conda is a **cross-platform package and environment management system** that plays two roles simultaneously: it installs packages (like pip) and manages isolated Python environments (like venv). More importantly, it is not limited to Python packages: C/C++ libraries, CUDA runtimes, and even R language packages all fall within its scope.
This guide covers installation, configuration, customization, and troubleshooting.
**Conda vs pip**
* **pip** primarily manages Python packages from PyPI and installs only what the package author has bundled. For C extensions, it uses a pre-built binary wheel if one is available; otherwise it compiles from source, relying on system-installed build tools and external C/C++ libraries.
* **conda** is an "environment + package manager." Its packages can bundle Python itself, C/C++ libraries, CUDA runtimes, and even compilers, and it uses a dependency-resolution mechanism to ensure the entire binary stack is mutually compatible. This makes it significantly more stable for scientific computing and ML/DL workloads; installing packages like `numpy`, `scipy`, and `pytorch` that depend heavily on BLAS/LAPACK/CUDA no longer requires managing low-level dependencies by hand.
***
## Anaconda vs Miniconda
The official website offers two options: **Anaconda** and **Miniconda**. Both include `conda` as their core tool; the difference is how much comes pre-bundled.
**Full bundle**: ships with 300+ pre-installed packages (NumPy, Pandas, Jupyter, Scikit-learn, ...), ready to use out of the box.
* Installation size: approximately **4.4 GB**
* Includes Anaconda Navigator (GUI management interface)
* Best for: **beginners** and those who prefer not to worry about package installation
> Personally recommended.
**Minimal installation**: only conda + Python + essential dependencies, giving you a clean starting point.
* Installation size: approximately **80 MB**
* CLI-only operation
* Best for: **experienced developers**, CI/CD pipelines, container environments, disk-space-constrained setups, and avoiding Anaconda licensing concerns
| Comparison | Anaconda | Miniconda |
| :------------------------ | :--------------------------------------------------- | :--------------------------------------- |
| Installation size | \~1.1 GB | \~90 MB |
| Pre-installed packages | 300+ data science packages | conda + Python only |
| GUI management | Anaconda Navigator | None |
| Use case | Quick start, teaching environments | Precise control, production environments |
| Customization flexibility | Lower (many pre-installed packages you may not need) | Maximum (install only what you need) |
**Which should I choose?**
**Plenty of disk space and want to get started quickly → Anaconda.**
**Want precise environment control, or need to conserve SSD space → Miniconda.** `Recommended`
In practice, most professional developers prefer Miniconda, because of the 300+ packages Anaconda pre-installs, typically only a few dozen are ever used.
***
## Installation
### Download Miniconda
Go to the [Miniconda official download page](https://www.anaconda.com/download) and select the **Windows 64-bit** Miniconda `.exe` installer.
Double-click the downloaded `Miniconda3-latest-Windows-x86_64.exe` and click **Next**.
Read and accept the License Agreement.
* **Just Me (recommended)**: Installs for the current user only; no administrator privileges required.
* **All Users**: System-wide installation; requires administrator privileges.
Unless you have a clear multi-user requirement, **Just Me** is the right choice.
The default path is `C:\Users\\miniconda3`.
To install to a different location (such as the D drive), change the path at this step. See the [Installing to the D Drive](#installing-to-the-d-drive) section below for details.
**The path must not contain spaces or non-ASCII characters**
Conda's installation path **cannot contain spaces or non-ASCII characters**. Paths like `D:\Program Files\conda` or `D:\MyApps\conda` will cause problems. Use a clean path such as `D:\miniconda3`.
The installer will ask about two options:
| Option | Recommendation | Description |
| :------------------------------------------ | :------------------ | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Create start menu shortcuts** | Check | Creates an Anaconda Prompt shortcut in the Start Menu |
| **Add Miniconda3 to my PATH** | **Leave unchecked** | Not recommended officially: conda's binary directory contains other package binaries; permanently adding it to PATH can conflict with other software. Use Anaconda Prompt or [`conda init`](#q-conda-command-not-found) instead |
| **Register Miniconda3 as default Python** | Situational | Check if there is no other Python on the system (checked by default) |
| **Clear the package cache upon completion** | Check | Runs `conda clean --all --force-pkgs-dirs` after installation to save space |
Click **Install**, wait for the installation to finish, then click **Finish**.
### Download Anaconda
Go to the [Anaconda official download page](https://www.anaconda.com/download) and download the Windows 64-bit installer.
Double-click `Anaconda3-20xx.xx-Windows-x86_64.exe` and follow the wizard.
Same as Miniconda: select **Just Me** and set the installation path.
**Disk space**
A full Anaconda installation requires approximately **4.4 GB** of disk space. With environments and package caches added later, it is advisable to have at least **10 GB** available.
Same as Miniconda: leave "Add to PATH" unchecked, and decide whether to register as the default Python based on your needs.
The installation process takes longer; just wait for it to finish.
### Silent command-line installation
For automated deployments or CI/CD environments, you can use the command line for a silent installation. The following commands are based on the [official Quickstart](https://www.anaconda.com/docs/getting-started/miniconda/install#quickstart-install-instructions):
**Silent installation implies acceptance of the Terms of Service**
Running a silent installation automatically accepts Anaconda's Terms of Service (TOS). Be sure to review the [Anaconda complete Terms of Service](https://anaconda.com/legal) before proceeding.
```powershell theme={null}
# Download the Miniconda installer
Invoke-WebRequest -Uri "https://repo.anaconda.com/miniconda/Miniconda3-latest-Windows-x86_64.exe" -OutFile ".\miniconda.exe"
```
```powershell theme={null}
# Silent install to `D:\miniconda3`
Start-Process -FilePath ".\miniconda.exe" -ArgumentList "/S /D=D:\miniconda3" -Wait
```
```powershell theme={null}
# Remove the installer
Remove-Item ".\miniconda.exe"
```
After installation, initialize conda:
```powershell theme={null}
# Initialize PowerShell support
D:\miniconda3\Scripts\conda.exe init powershell
# Changes take effect after reopening PowerShell
```
```cmd theme={null}
:: Download the Miniconda installer
curl https://repo.anaconda.com/miniconda/Miniconda3-latest-Windows-x86_64.exe -o .\miniconda.exe
:: Silent install to D:\miniconda3
start /wait "" .\miniconda.exe /S /D=D:\miniconda3
:: Remove the installer
del .\miniconda.exe
```
After installation, open [Anaconda Prompt](https://www.anaconda.com/docs/reference/glossary#anaconda-prompt) to use conda, or run the following to initialize CMD support:
```cmd theme={null}
:: Initialize CMD support
D:\miniconda3\Scripts\conda.exe init cmd.exe
:: Changes take effect after reopening CMD
```
```bash theme={null}
# Create the installation directory
mkdir -p ~/miniconda3
# Download the installer script (Apple Silicon)
curl https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-arm64.sh -o ~/miniconda3/miniconda.sh
# Silent install
bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
# Remove the installer script
rm ~/miniconda3/miniconda.sh
```
**Intel Mac**
If you are using an Intel chip, replace `arm64` with `x86_64` in the download URL.
After installation, initialize conda:
```bash theme={null}
# Activate conda
source ~/miniconda3/bin/activate
# Initialize all shells
conda init --all
```
```bash theme={null}
# Create the installation directory
mkdir -p ~/miniconda3
# Download the installer script (x86_64)
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda3/miniconda.sh
# Silent install
bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
# Remove the installer script
rm ~/miniconda3/miniconda.sh
```
**ARM64 / AWS Graviton**
If you are using an ARM64 architecture, replace `x86_64` with `aarch64` in the download URL.
After installation, initialize conda:
```bash theme={null}
# Activate conda
source ~/miniconda3/bin/activate
# Initialize all shells
conda init --all
```
For quick commands, see [Installing to the D Drive](#installing-to-the-d-drive) (Windows).
**Parameter notes**
**Windows**:
* `/S`: Silent mode (no GUI)
* `/D=`: Specifies the installation path; **must be the last argument**, and the path **must not** be wrapped in quotes
* Without `/D=`, the default installation path is `C:\Users\\miniconda3`
**macOS / Linux**:
* `-b`: Batch mode (silent install, no interactive confirmation)
* `-u`: Update mode; overwrites if the path already exists
* `-p`: Specifies the installation path
***
## Post-installation Verification
After installation, open **PowerShell** or **Anaconda Prompt** (search from the Start Menu) and run the following verification commands:
> If `conda` is not recognized, see [Troubleshooting](#q-conda-command-not-found).
```powershell theme={null}
# Check the conda version
conda --version
# Show full environment information
conda info
```
Expected output example:
```text theme={null}
conda 24.x.x
active environment : base
active env location : D:\miniconda3
user config file : C:\Users\yourname\.condarc
populated config files : D:\miniconda3\.condarc
conda version : 24.x.x
platform : win-64
...
```
There are two easily confused configuration file fields in the output. `.condarc` (conda run configuration) is conda's configuration file, written in YAML syntax. It controls channels, default paths, behavioral preferences, and all other conda settings. Conda allows multiple `.condarc` files to coexist at different locations:
| Field | Description |
| :------------------------- | :---------------------------------------------------------------------------------------------------------------------------------- |
| **user config file** | The user-level `.condarc`, always located at `C:\Users\\.condarc`. This path is shown even if the file does not yet exist |
| **populated config files** | `.condarc` files that **actually contain content**. When installed on the D drive, this is typically `D:\miniconda3\.condarc` |
**Which .condarc takes effect?**
Conda **reads all** `.condarc` files it finds simultaneously and merges their settings. If two files have conflicting settings, the user-level file (`C:\Users\...`) takes priority.
It is generally best to maintain only one `.condarc` to avoid scattered configuration. You can edit the one under the installation directory (`D:\miniconda3\.condarc`) directly, or use `conda config` commands (which write to the user-level file by default):
```powershell theme={null}
# View all active settings (merged result)
conda config --show
# Set a configuration value (writes to user-level .condarc)
conda config --set auto_activate_base false
# Show the source of each configuration setting
conda config --show-sources
```
### Create a test environment
Verify that conda can correctly create an isolated environment and install packages:
```powershell theme={null}
# Create a Python 3.12 test environment
conda create -n test_env python=3.12 -y
# Activate the environment
conda activate test_env
# Confirm the Python version
python --version
# Install a package to verify
conda install numpy -y
# Quick test
python -c "import numpy; print(f'NumPy {numpy.__version__} OK')"
# Clean up after testing
conda deactivate
conda remove -n test_env --all -y
```
**You should see the `(test_env)` prefix**
After successfully running `conda activate test_env`, the command prompt prefix changes from `(base)` to `(test_env)`, indicating that you are working inside that isolated environment.
***
## Custom Configuration
### Installing to the D Drive
#### Why install to the D drive?
Many Windows users have a disk layout like this:
| Drive | Purpose | Characteristics |
| :-------------------- | :----------------------------------- | :---------------------------------- |
| **C drive** (SSD) | Operating system + core applications | Smaller capacity (128–512 GB), fast |
| **D drive** (HDD/SSD) | Data and development tools | Larger capacity (1 TB+) |
Conda environments and package caches grow continuously. A single ML project environment can easily reach **2–5 GB** (especially with PyTorch or TensorFlow), and multiple projects will quickly consume C drive space. Installing Conda to the D drive lets you:
* Avoid system instability caused by a full C drive
* Preserve development environments when reinstalling the OS
* Keep environments separate from the system for cleaner maintenance
**Performance trade-off**
If your D drive is a traditional HDD (hard disk drive), conda package extraction and environment creation will be noticeably slower than on an SSD.
If the D drive is also an SSD, this concern does not apply.
#### How to do it
Use the `/D=` parameter to specify the path (you can pre-create a miniconda3 folder on the D drive):
```powershell theme={null}
Start-Process -FilePath ".\miniconda.exe" -ArgumentList "/S /D=D:\miniconda3" -Wait
```
In the installer wizard's "Choose Install Location" step, change the path to `D:\miniconda3`.
The directory structure after installing to the D drive looks like this:
**Finding your environments**
All environments created with `conda create -n myenv` are stored by default under `miniconda3/envs/`. You can reference an environment's Python interpreter directly, for example `D:\miniconda3\envs\ml_project\python.exe`.
#### Advanced: custom environment and cache paths
If you want to store environments and caches outside the Miniconda3 directory (for example, to share environments across multiple Conda installations or for finer-grained disk management), you can configure this in the [`.condarc` configuration file](#condarc-configuration-file):
```yaml title=".condarc" theme={null}
# Custom environment storage path
envs_dirs:
- D:\conda_envs
# Custom package cache path
pkgs_dirs:
- D:\conda_pkgs
```
**.condarc location**
The `.condarc` file lives in the user home directory: `C:\Users\\.condarc`. If it does not exist, you can create it manually, or run `conda config --set envs_dirs D:\conda_envs` to have conda create it automatically.
***
### The Channel Mechanism
#### What is a channel?
A channel is a **source repository** for conda packages: essentially a remote directory (URL) containing pre-compiled packages. When you run `conda install numpy`, conda searches channels in priority order and downloads the first matching package version it finds.
Think of it like an app store for your phone:
| Concept | Analogy |
| :------------------- | :-------------------------------- |
| **Channel** | App Store (package source) |
| **conda install** | Downloading an app from the store |
| **Channel priority** | Which store to search first |
#### Major channels
| Channel | Maintainer | Highlights | License |
| :-------------- | :------------------- | :----------------------------------------- | :-------------------------------------------- |
| **defaults** | Anaconda, Inc. | Stable, the default channel | May require a paid license for commercial use |
| **conda-forge** | Community-maintained | Largest package selection, fastest updates | **Completely free** |
| **pytorch** | PyTorch team | Official latest PyTorch releases | Free |
| **nvidia** | NVIDIA | CUDA toolkit, cuDNN | Free |
**Anaconda Terms of Service**
Since 2024, Anaconda has updated the Terms of Service for the **defaults channel** and **Anaconda Distribution**:
* **Individual use and organizations with fewer than 200 employees**: Free to use;
* **Organizations with 200 or more employees or contractors**: A paid license is required to access the defaults channel or Anaconda Distribution.
* **Educational institutions**: Pure classroom use is generally free, but large universities or research institutions conducting research may be considered "organizational use" and may not be fully exempt.
If you are in a large enterprise or institution and only need the conda ecosystem without depending on the Anaconda defaults channel, the recommended approach is **Miniconda/Mambaforge** with **conda-forge** as the primary channel: fully open source, free, and not subject to Anaconda's licensing terms. conda-forge is community-maintained, completely free for commercial use, and unaffected by Anaconda's fee structure.
#### Configuring channels
```powershell theme={null}
# View current channel configuration
conda config --show channels
# Add conda-forge as the highest priority
conda config --add channels conda-forge
# Enable strict priority mode (recommended)
conda config --set channel_priority strict
# Specify a channel for a single install (one-time)
conda install -c pytorch pytorch torchvision
```
**What is strict priority mode?**
```text theme={null}
Channel list:
1. conda-forge ← Highest priority, searched first
2. defaults ← Searched only if conda-forge doesn't have it
In strict mode: if conda-forge has numpy 1.26 and defaults has numpy 1.27,
conda still installs conda-forge's 1.26, because it comes from the higher-priority channel.
```
**Recommended .condarc channel configuration**
```yaml theme={null}
channels:
- conda-forge
- defaults
channel_priority: strict
```
With this configuration, conda-forge is the primary source (free and fast-updating), with defaults as a fallback. See the full example in the [`.condarc` configuration file](#condarc-configuration-file) section.
***
### .condarc Configuration File
`.condarc` is conda's global configuration file, written in YAML syntax. The following is a complete example of commonly used settings:
```yaml title=".condarc" theme={null}
# ── Channel settings ──
channels:
- conda-forge
- defaults
channel_priority: strict
# ── Path settings ──
envs_dirs:
- D:\conda_envs
pkgs_dirs:
- D:\conda_pkgs
# ── Behavior settings ──
auto_activate_base: false # Whether to auto-activate the base environment (recommended: false)
auto_update_conda: true # Whether to auto-update conda
show_channel_urls: true # Show the source channel when installing packages
```
**auto\_activate\_base: false**
By default, opening a new terminal activates the `(base)` environment. Setting this to `false` requires a manual `conda activate base` to enter it. This prevents accidental pollution of the base environment and is a good practice.
***
## Quick Reference
### Environment management
```powershell theme={null}
# Create an environment (specify Python version)
conda create -n myenv python=3.12
# Create an environment at a specific path (use --prefix instead of -n)
{/* conda create python=3.9 --prefix D:\Miniconda3\envs\python39 -y */}
# Create an environment from a YAML file
conda env create -f environment.yml
# Activate / deactivate an environment
conda activate myenv
conda deactivate
# List all environments
conda env list
# Remove an environment
conda remove -n myenv --all
# Export the environment (reproducible snapshot)
conda env export > environment.yml
# Export only manually installed packages (more cross-platform compatible)
conda env export --from-history > environment.yml
```
### Package management
```powershell theme={null}
# Install packages
conda install numpy pandas matplotlib
# Install a specific version
conda install python=3.11 numpy=1.26
# Install from a specific channel
conda install -c conda-forge scikit-learn
# Update a package
conda update numpy
# Update all packages
conda update --all
# Search for a package
conda search pytorch
# Remove a package
conda remove numpy
# List installed packages
conda list
```
**Notes on mixing conda and pip**
You can use pip to install packages not available in conda channels, but keep these points in mind:
1. **conda first, pip second**: Install what you can with conda first, then fill in the gaps with pip.
2. **Avoid repeatedly alternating**: conda does not track pip-installed packages; alternating between the two repeatedly tends to cause dependency conflicts.
3. **Watch the export**: `conda env export` records both conda-installed and pip-installed packages.
***
## Troubleshooting
### Q: `conda` command not found?
**A:** If you did not check "Add to PATH" during installation, a regular CMD or PowerShell session will not recognize the `conda` command.
There are two solutions:
**Method 1: Use `conda init` (recommended)**
Run this once from Anaconda Prompt; after that, all PowerShell sessions will recognize conda:
```powershell theme={null}
D:\miniconda3\Scripts\conda.exe init powershell
```
Changes take effect after reopening PowerShell.
**Method 2: Manually add conda to the PATH environment variable**
If `conda init` does not work correctly, add conda's paths to the system PATH manually:
Type "environment variables" in the Windows search bar and open "Edit the system environment variables".
Click the "Environment Variables" button.
Find **Path** under "User variables" or "System variables" and click "Edit".
Click "New" and add the following paths in order (adjust to match your actual installation location):
```text theme={null}
D:\Miniconda3
D:\Miniconda3\Library\bin
D:\Miniconda3\Scripts
D:\Miniconda3\condabin
```
**Anaconda users**
Replace `Miniconda3` with `Anaconda3` in the paths above.
Click "OK" repeatedly to close all windows.
**Reopen** the terminal and run `conda --version` to verify.
### Q: `CommandNotFoundError` after activating an environment?
**A:** This is commonly caused by PowerShell's execution policy blocking conda's initialization script. Run:
```powershell theme={null}
Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser
```
Then reopen PowerShell.
### Q: `conda install` is extremely slow at dependency resolution (Solving environment)?
**A:** This is a classic conda pain point. Solutions:
1. **Use the libmamba solver** (built in for conda >= 22.11):
```powershell theme={null}
conda config --set solver libmamba
```
2. **Set strict [channel priority](#configuring-channels)** (reduces the search scope):
```powershell theme={null}
conda config --set channel_priority strict
```
3. **Avoid installing packages in the base environment**: Keep base clean and create a separate environment for each project.
### Q: `CondaHTTPError` during installation or update?
**A:** This is usually a network or proxy configuration issue. Set a proxy in [`.condarc`](#condarc-configuration-file):
```yaml title=".condarc" theme={null}
proxy_servers:
http: http://proxy.example.com:8080
https: https://proxy.example.com:8080
```
### Q: Can Anaconda and Miniconda be installed at the same time?
**A:** Technically yes, but it is **strongly not recommended**. Both share the same `conda` core, and having both installed causes PATH and environment variable conflicts. Choose one.
### Q: How do I uninstall?
**A:** Refer to the official uninstall guides:
* [Uninstall Miniconda](https://www.anaconda.com/docs/getting-started/miniconda/uninstall#uninstall-procedure)
* [Uninstall Anaconda Distribution](https://www.anaconda.com/docs/getting-started/anaconda/uninstall#uninstall-procedure)
***
## Related Pages
* [Windows CUDA Development Environment Setup Guide](/en/notes/ai-core/environment/cuda-setup/)
* [Miniconda Official Installation Guide](https://www.anaconda.com/docs/getting-started/miniconda/install)
* [Anaconda Official Installation Guide](https://www.anaconda.com/docs/getting-started/anaconda/install)