Update instructions.

README.md

@@ -6,9 +6,36 @@ results.
 
 The concept is to collect both results simultaneously during a training phase and use it to learn a model that may be used later, under the same conditions, but without the high-resolution, invasive spectrometer. The idea is that minimum tuning of the parameters of this methods are needed, so that if the data for training is available, no fine-tuning is required.
 
-## Installation
+## Installation in Maxwell
 
-One may install it simply with `pip install pes_to_spec`.
+The dependencies can be installed in a separate environment and used from Juyter Lab in Maxwell, by executing the following commands in Maxwell:
+
+```
+# copy it
+git clone https://git.xfel.eu/machineLearning/pes_to_spec.git
+cd pes_to_spec
+
+# create separate environment
+python -m venv env
+source env/bin/activate
+
+# install dependencies
+pip install --force-reinstall -i https://pypi.anaconda.org/intel/simple --no-dependencies numpy scipy
+pip install numpy scipy
+pip install torch --index-url https://download.pytorch.org/whl/cpu
+
+# add it to the Jupyter Lab in Maxwell
+pip install ipykernel
+python -m ipykernel install --name pes_to_spec_env
+
+```
+
+After this, open the example notebook from the `notebook` directory in a Jupyter Lab instance (max-jhub.desy.de) and set the kernel to `pes_to_spec_env`.
+
+## Installation in an existing environment
+
+If one just wants to install it standalone, simply type:
+`pip install pes_to_spec`
 
 While the dependencies should be automatically used, using the Intel-optimized `numpy` and `scipy` packages is recommended, as they are much faster.
 This has been tested with `scipy==1.7.3`, but it should work with any version of scipy and numpy.