app_.py

import streamlit as st
import pandas as pd
import matplotlib.pyplot as plt
import requests
import io
from transformers import pipeline
from PIL import Image

#import streamlit as st
#import pandas as pd
#import matplotlib.pyplot as plt



st.title('Playing cards Image Analysis')


#sample slider; feel free to remove:
#x = st.slider('Select a value')
#st.write(x, 'squared is', x * x)

'''
This next piece of code will hit GitHub for two csv files
One is the original dataset, broken up into test, train, valid.
The second csv is the test dataset, with the results after the models were run through the API
'''
# Downloading the csv file from your GitHub account
url = "https://huggingface.co/datasets/rwcuffney/autotrain-data-pick_a_card/raw/main/cards.csv" 
download = requests.get(url).content

# Reading the downloaded content and turning it into a pandas data frame
df = pd.read_csv(io.StringIO(download.decode('utf-8')))
#df = pd.read_csv('playing_cards/cards.csv').sort_values('class index')
df_fulldataset=df

# Downloading the csv file from your GitHub account
url = "https://huggingface.co/datasets/rwcuffney/autotrain-data-pick_a_card/raw/main/ML_results.csv" 
download = requests.get(url).content

# Reading the downloaded content and turning it into a pandas data frame
df = pd.read_csv(io.StringIO(download.decode('utf-8')))
#df = pd.read_csv('playing_cards/cards.csv').sort_values('class index')
df_test = df



# Create the button
if st.button('Click me to re-run code',key='RunCode_button'):
    # Call the function when the button is clicked
    st.experimental_rerun()

st.header('Sample of the .csv data:')
x = st.slider('Select a value',value=10,max_value=8000)
st.table(df_fulldataset.sample(x))

### HORIZONTAL BAR ###

st.header('Distribution of the playing card images:')

# Get the value counts of the 'labels' column
value_counts = df_fulldataset.groupby('labels')['class index'].count().iloc[::-1]


fig, ax = plt.subplots(figsize=(10,10))
    
# Create a bar chart of the value counts
ax = value_counts.plot.barh()
# Set the chart title and axis labels
ax.set_title('Value Counts of Labels')
ax.set_xlabel('Label')
ax.set_ylabel('Count')

# Show the chart
st.pyplot(fig)


### PIE CHART ###

st.header('Balance of Train,Valid,Test datasets:')

# Get the value counts of the 'labels' column
value_counts = df_fulldataset.groupby('data set')['class index'].count().iloc[::-1]

value_counts =df_fulldataset['data set'].value_counts()

fig, ax = plt.subplots(figsize=(5,5)
                      )
# Create a bar chart of the value counts
ax = value_counts.plot.pie(autopct='%1.1f%%')

# Set the chart title and axis labels
# Show the chart
st.pyplot(fig)





models_run= ['SwinForImageClassification_24', 
             'ViTForImageClassification_22',
             'SwinForImageClassification_21', 
             'ResNetForImageClassification_23',
             'BeitForImageClassification_25']


from enum import Enum
 
API_dict = dict(
    SwinForImageClassification_21="https://api-inference.huggingface.co/models/rwcuffney/autotrain-pick_a_card-3726099221",
    ViTForImageClassification_22="https://api-inference.huggingface.co/models/rwcuffney/autotrain-pick_a_card-3726099222",
    ResNetForImageClassification_23= "https://api-inference.huggingface.co/models/rwcuffney/autotrain-pick_a_card-3726099223",
    SwinForImageClassification_24 = "https://api-inference.huggingface.co/models/rwcuffney/autotrain-pick_a_card-3726099224",
    BeitForImageClassification_25="https://api-inference.huggingface.co/models/rwcuffney/autotrain-pick_a_card-3726099225")

pipeline_dict = dict(
    SwinForImageClassification_21="rwcuffney/autotrain-pick_a_card-3726099221",
    ViTForImageClassification_22="rwcuffney/autotrain-pick_a_card-3726099222",
    ResNetForImageClassification_23= "rwcuffney/autotrain-pick_a_card-3726099223",
    SwinForImageClassification_24 = "rwcuffney/autotrain-pick_a_card-3726099224",
    BeitForImageClassification_25="rwcuffney/autotrain-pick_a_card-3726099225")


# printing enum member as string
#print(Api_URL.ViTForImageClassification_22.value)


####Try it out ###
import requests

st.header("Try it out")

'''
Warning: it will error out at first, resubmit a few times.
Each model needs to 'warm up' before they start working.

You can use any image... try test/queen of hearts/4.jpg to see an example that 
Got different results with different models
'''

headers = {"Authorization": "Bearer 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"}
def query(filename,api_url):
    #with open(filename, "rb") as f:
        #data = f.read()
    response = requests.post(api_url, headers=headers, data=filename)
    return response.json()

#API_URL = "https://api-inference.huggingface.co/models/rwcuffney/autotrain-pick_a_card-3726099224"





##### FORM #####

with st.form("api_form"):
    api = st.selectbox('Which model do you want to try?',models_run,key='select_box')




    uploaded_file = st.file_uploader("Choose a file")
    if uploaded_file is not None:
        # To read file as bytes:
        bytes_data = uploaded_file.getvalue()
        #st.write(bytes_data)
        st.image(uploaded_file)
        image = Image.open(uploaded_file)


    submitted = st.form_submit_button("Submit")
    if submitted:
        pipeline = pipeline(task="image-classification", model=pipeline_dict[api])
        def predict(image):
            predictions = pipeline(image)
            return {p["label"]: p["score"] for p in predictions}
        prediction = predict(image)
        st.write(prediction)
        #st.write(API_dict[api])
        #output = query(bytes_data,API_dict[api])

        #prediction = output[0]['label']
        #st.write(f'prediction = {prediction}')
        #st.text(output)
        




#### FUNCTIONS ####
import sklearn
from sklearn import metrics
import matplotlib.pyplot as plt

index = ['accuracy_score','Weighted f1', 'Cohen Kappa','Matthews']
df_Metrics =pd.DataFrame(index=index)

labels = df_test['labels'].unique()



### FUNCTION TO SHOW THE METRICS
def show_metrics(test,pred,name):
    from sklearn import metrics
    
    my_Name = name
    my_Accuracy_score=metrics.accuracy_score(test, pred)
    #my_ROC_AUC_score= roc_auc_score(y, model.predict_proba(X), multi_class='ovr')
    my_Weighted_f1= metrics.f1_score(test, pred,average='weighted')
    my_Cohen_Kappa = metrics.cohen_kappa_score(test, pred)
    my_Matthews_coefficient=metrics.matthews_corrcoef(test, pred)
    
    st.header(f'Metrics for {my_Name}:')	
    report =metrics.classification_report(test, pred, output_dict=True)
    df_report = pd.DataFrame(report).transpose()	
    st.dataframe(df_report )
    st.write(f'Accuracy Score........{metrics.accuracy_score(test, pred):.4f}\n\n' \
          #f'ROC AUC Score.........{my_ROC_AUC_score:.4f}\n\n' \
          f'Weighted f1 score.....{my_Weighted_f1:.4f}\n\n' \
          f'Cohen Kappa...........{my_Cohen_Kappa:.4f}\n\n' \
          f'Matthews Coefficient..{my_Matthews_coefficient:.4f}\n\n')
    my_List = [my_Accuracy_score, my_Weighted_f1, my_Cohen_Kappa, my_Matthews_coefficient]

    df_Metrics[my_Name] = my_List
    
    cfm= metrics.confusion_matrix(test, pred)
    st.caption(f'Confusion Matrix: {my_Name}')
    cmd = metrics.ConfusionMatrixDisplay(cfm,display_labels=labels)
    fig, ax = plt.subplots(figsize=(15,15))
    ax = cmd.plot(ax=ax, 
                  colorbar=False,
                  values_format = '.0f',
                  cmap='Reds')#='tab20')# see color options here https://matplotlib.org/stable/tutorials/colors/colormaps.html
    plt.xticks(rotation=90)
    st.pyplot(fig)





st.header('Let\'s see how the models performed')

'''
The next part of the code will analyze the full dataset.
Choose all five models to compare them all

'''


##### FORM #####

with st.form("my_form"):
    st.write("You can choose from 1 to 5 models")


    selected_options = st.multiselect(
    'Which models would you like to analyze?', models_run)

    submitted = st.form_submit_button("Submit")
    if submitted:
        st.write('you selected',selected_options)


    ###Show the metrics for each dataset:
        test = df_test['labels']

        #for m in models_run:
        for m in selected_options:
            pred = df_test[m]
            show_metrics(test,pred,m)

        st.header('Metrics for all models:')
        st.table(df_Metrics)

        #### GRAPH THE RESULTS ###
        import seaborn as sns

        # Reshape the dataframe into long format using pd.melt()
        #subset_df = pd.melt(df_Metrics[['SwinForImageClassification_24', 
        #'ViTForImageClassification_22', 'SwinForImageClassification_21', 'ResNetForImageClassification_23', 'BeitForImageClassification_25']].reset_index(), id_vars='index', var_name='Model', value_name='Score')
        subset_df = pd.melt(df_Metrics[selected_options].reset_index(), id_vars='index', var_name='Model', value_name='Score')

        sns.set_style('whitegrid')
        ax=sns.catplot(data=subset_df, 
                    x='index', 
                    y='Score', 
                    hue='Model', 
                    kind='bar', 
                    palette='Blues', 
                    aspect=2)

        plt.xlabel('Clusters')
        plt.ylabel('Scores')

        fig = ax.figure
        st.pyplot(fig)