Limitations on package installation (indsl library) when making Streamlit apps in CDF

Hi Sangavi,

As you've noted, importing InDSL into Pyodide-based environments (like Streamlit) is challenging due to the dependencies that are not readily available on Pyodide. We're aware of this limitation and understand the inconvenience it causes.

Your proposed solution to use Cognite Functions to handle computations with the InDSL library and then use the results in Streamlit for dashboarding is a valid approach. This method utilizes Cognite Functions as a backend service, where you can perform your calculations and then visualize the results in Streamlit.

That being said, there are some workarounds to bypass the import limitations directly within Streamlit, although they are not very elegant. Here’s an example code snippet that manages to import indsl into streamlit (you’ll have to add plotly as a dependency before trying this out):

import micropip
await micropip.install("indsl==8.2.1", deps=False)
await micropip.install("https://files.pythonhosted.org/packages/24/21/7d397a4b7934ff4028987914ac1044d3b7d52712f30e2ac7a2ae5bc86dd0/typing_extensions-4.8.0-py3-none-any.whl")
await micropip.install("typeguard", deps=False)
await micropip.install("scipy")
await micropip.install("matplotlib")
await micropip.install("fluids")

import streamlit as st
import plotly.graph_objects as go
from cognite.client import CogniteClient
import indsl

st.title("InDSL in Streamlit")
client = CogniteClient()

ts_list = client.time_series.list(
    external_id_prefix="EVE",
    unit="C",
    limit=5
)
ts = st.selectbox(
    label="Select a time series",
    options=ts_list,
    format_func=lambda x: f"{x.external_id} - {x.name}"
)

dps = client.time_series.data.retrieve_dataframe(
    external_id=ts.external_id,
    start="2w-ago",
    end="now",
    aggregates=["average"],
    granularity="1h",
    include_aggregate_name=False,
    limit=None
)

# run an InDSL method
dps_smooth = indsl.smooth.sg(data=dps.iloc[:,0])

fig = go.Figure(
    data=[
        go.Scatter(
            x=dps.index,
            y=dps.iloc[:,0],
            name="Sensor data",
            mode="lines",
            showlegend=True
        ),
        go.Scatter(
            x=dps_smooth.index,
            y=dps_smooth.values,
            name="Smoothed data",
            mode="lines",
            showlegend=True
        ),
    ],
    layout={
        "xaxis": {"showspikes": True, "spikemode": "across"},
        "legend": {"orientation": "h"},
        "hovermode": "x",
        "spikedistance": -1,
        "margin": {"l": 20, "r": 20, "b": 10, "t": 20, "pad": 0},
        "height": 600
    }
)

st.plotly_chart(fig, use_container_width=True)

You might need to adapt the time series filter to the particular CDF project where you are testing this in. Here's a screenshot of how this looks like:

I hope this information helps with your use case.