uploaded files

app.py

@@ -0,0 +1,121 @@
+import streamlit as st
+import pandas as pd
+import os
+from utils import save_json, load_json
+from markdown import markdown
+from utils import load_json
+from autoviz import AutoViz_Class
+import base64
+from google.cloud import aiplatform
+import base64
+import vertexai
+from vertexai.preview.generative_models import GenerativeModel, Part
+
+#setup cloud
+aiplatform.init(
+    project = "ultra-heading-407815",
+    location="us-central1"
+    )
+
+os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "credentials.json"
+
+
+dataframe = None
+st.title("GemInsights")
+file = st.file_uploader(
+    "Pick a dataframe", type=["csv", "xlsx"], accept_multiple_files=False
+)
+
+if file is not None:
+    _, extension = os.path.splitext(file.name)
+    if extension == ".csv":
+        dataframe = pd.read_csv(file)
+    else:
+        dataframe = pd.read_excel(file)
+    st.write(dataframe.head())
+    st.write(f"updated a dataframe with shape {dataframe.shape}")
+
+if file is not None:
+    text_input = st.text_input(
+        "Enter something about the data 👇",
+        label_visibility="visible",
+        disabled=False,
+        placeholder="eg:- This is a sales dataframe",
+    )
+
+    option = st.selectbox(
+        "Which is the target column?",
+        tuple(list(dataframe.columns)),
+        index=None,
+        placeholder="Select one column in here",
+    )
+
+def plot(dataframe, target):
+
+    AV = AutoViz_Class()
+
+    dft = AV.AutoViz(
+    "",
+    sep=",",
+    depVar=target,
+    dfte=dataframe,
+    header=0,
+    verbose=2,
+    lowess=False,
+    chart_format="jpg",
+    max_rows_analyzed=500,
+    max_cols_analyzed=20,
+    save_plot_dir="plots",
+    )
+
+def prompt_make(dataframe, target, info):
+    images = []
+    image_dir = f"plots/{target}"
+    image_files = os.listdir(image_dir)
+    for image_file in image_files:
+        image_path = os.path.join(image_dir, image_file)
+        img = open(image_path, "rb").read()
+        img_bytes = Part.from_data(
+            base64.b64decode(base64.encodebytes(img)), mime_type="image/jpeg"
+        )
+        images.append(img_bytes)
+    with open("prompt.txt", "rb") as file:
+        data = file.read()
+    prompt = f"{data}\n Here are some of the informations related to the dataset - '{info}'"
+    
+    # print(f"{prompt}")
+    # print(images)
+    return prompt, images
+
+def generate_res(prompt, images):
+    print("prompting ...")
+    model = GenerativeModel("gemini-pro-vision")
+    responses = model.generate_content(
+        [prompt]+images,
+        generation_config={
+            "max_output_tokens": 2048,
+            "temperature": 0.4,
+            "top_p": 1,
+            "top_k": 32
+        },
+    )
+    return responses.text
+
+
+
+def generate(dataframe, text_input, option):
+    plot(dataframe, option)
+    prompt, images = prompt_make(dataframe, option, text_input)
+    res = generate_res(prompt, images)
+    return res
+
+if st.button("Get Insights", type="primary"):
+    st.write("generating insights ...")
+    # running the pipeline
+
+    response = generate(dataframe, text_input, option)
+    res = markdown(response)
+    st.markdown(res, unsafe_allow_html=True)
+
+else:
+    st.write("")

prompt.txt

@@ -0,0 +1,26 @@
+Act as an intelligent data Analyst who communicates in simple English and clear messages to the clients
+
+We build an end-to-end application that involves visualizing datasets, and we aim to extract valuable insights from these visualizations using llm. The insights generated should be beneficial to both companies and end-users. It's crucial that the model refrains from explicitly mentioning the images and provides information in a clear, detailed, and actionable manner.
+Here are the visualizations of the data. give the insights by considering the following points
+
+Here are important notes for output generation:
+- Analyze the visual elements within the dataset using the visualizations.
+- Identify and describe any prominent trends, patterns, or anomalies observed in the visual representations.
+- Derive insights that are specifically relevant to the industry or domain associated with the dataset.
+- Emphasize actionable information that could be of value to companies operating in that industry.
+- Explore the possibility of making predictions based on the visual content.
+- Formulate insights that would be valuable from an end-user perspective.
+- Consider how the extracted information can enhance user experience, decision-making, or engagement.
+- Do not mention the images directly in your responses. Focus on conveying insights without explicitly stating the visual content.
+- Ensure that the insights are presented in a language suitable for technical and non-technical audiences. I encourage you to give clear, detailed explanations.
+- Prioritize insights that are actionable and can contribute to informed decision-making for both businesses and end-users.
+- If there are any recognized design patterns or industry standards applicable to the analysis, please incorporate and explain them.
+
+Note to Model:
+- Do not explicitly reference the images in your responses.
+- Focus on providing clear, detailed, and actionable insights.
+- Ensure that the insights are presented in a language suitable for technical and non-technical audiences.
+
+Remember to adapt the prompt based on the specific details of your dataset and the objectives of your application.
+Give important actionable insights rather than giving all. give as pointwise. don't mention the visualizations of plots in the output.
+don't use too many statistics jargons also. 

requirements.txt

@@ -0,0 +1,16 @@
+google-generativeai
+pandas
+numpy
+matplotlib
+seaborn
+python-box
+pexpect
+streamlit
+dataframe_image
+jinja2
+PyYAML
+autoviz
+ipython
+google-cloud-aiplatform
+markdown
+llama-index

utils.py

@@ -0,0 +1,11 @@
+import json
+from box import ConfigBox
+
+def load_json(file):
+    with open(path) as f:
+        content = json.load(f)
+    return ConfigBox(content)
+
+def save_json(file, content):
+    with open(path, "w") as f:
+        json.dump(data, f, indent=4)