Fix docs about installation (#163)

README.md

@@ -22,15 +22,17 @@ TorchANI is a pytorch implementation of ANI. It is currently under alpha release
 
 # Install
 
-TorchANI requires the latest preview version of PyTorch. You can install PyTorch by
+TorchANI requires the latest preview version of PyTorch. You can install PyTorch by the following commands (assuming cuda10):
 
 ```bash
-conda install pytorch-nightly -c pytorch
+pip install numpy torchvision_nightly
+pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cu100/torch_nightly.html
 ```
+
 If you updated TorchANI, you may also need to update PyTorch:
 
 ```bash
-conda update pytorch-nightly -c pytorch
+pip install --upgrade torch_nightly -f https://download.pytorch.org/whl/nightly/cu100/torch_nightly.html
 ```
 
 After installing the correct PyTorch, you can install TorchANI by:
@@ -41,6 +43,8 @@ pip install torchani
 
 See also [PyTorch's official site](https://pytorch.org/get-started/locally/) for instructions of installing latest preview version of PyTorch.
 
+Please install nightly PyTorch through `pip install` instead of `conda install`. If your PyTorch is installed through `conda install`, then `pip` would mistakenly recognize the package name as `torch` instead of `torch-nightly`, which would cause dependency issue when installing TorchANI.
+
 # Paper
 
 The original ANI-1 paper is:

docs/api.rst

@@ -68,5 +68,4 @@ Ignite Helpers
 .. autofunction:: torchani.ignite.MSELoss
 .. autoclass:: torchani.ignite.DictMetric
 .. autofunction:: torchani.ignite.RMSEMetric
-.. autoclass:: torchani.ignite.MaxAbsoluteError
 .. autofunction:: torchani.ignite.MAEMetric

docs/start.rst

 Installation
 ============
 
-TorchANI requires the latest preview version of PyTorch. If you are using anaconda, you can install PyTorch by ``conda install pytorch-nightly -c pytorch``. You can also refer to `the official site of PyTorch`_ for instructions of installing latest preview version of PyTorch.
+TorchANI requires the latest preview version of PyTorch. You can install PyTorch by the following commands (assuming cuda10):
 
-After installing PyTorch, you can then install TorchANI by ``pip install torchani``.
+.. code-block:: bash
 
-.. _the official site of PyTorch:
+    pip install numpy torchvision_nightly
+    pip install torch_nightly -f https://download.pytorch.org/whl/nightly/cu100/torch_nightly.html
+
+If you updated TorchANI, you may also need to update PyTorch:
+
+.. code-block:: bash
+
+    pip install --upgrade torch_nightly -f https://download.pytorch.org/whl/nightly/cu100/torch_nightly.html
+
+
+After installing the correct PyTorch, you can install TorchANI by:
+
+.. code-block:: bash
+
+    pip install torchani
+
+See also `PyTorch's official site`_ for instructions of installing latest preview version of PyTorch.
+
+.. warning::
+
+    Please install nightly PyTorch through ``pip install`` instead of ``conda install``. If your PyTorch is installed through ``conda install``, then `pip` would mistakenly recognize the package name as `torch` instead of `torch-nightly`, which would cause dependency issue when installing TorchANI.
+
+.. _PyTorch's official site:
     https://pytorch.org/get-started/locally/