import pandas as pd import numpy as np import matplotlib.pyplot as plt import seaborn as sns

df = pd.read_csv('/work/tord_v3_edited.csv') df.info()

df.head(20)

df.describe()

# sanitize "raised_usd" df['raised_usd'] = df['raised_usd'].fillna(0) # create outcome variable success df['success'] = np.where(df['raised_usd'] >= 500000, 1, 0)

df['success'].describe()

sns.pairplot(df, vars = ['raised_usd','teamsize','success'])

# Potential multicollinearity issue corr = df.corr() sns.heatmap(corr) plt.show()

df['rating'] = df['rating'].fillna(0) df['ERC20'] = df['ERC20'].fillna(0) df['teamsize'] = df['teamsize'].fillna(1)

candidate_variables = ['teamsize','rating','bonus','sold_tokens','bounty','kyc'] current_variables = [] target = ['success'] df['ratxteamsize']=df['teamsize'] * df['rating'] df['ratxteamsize'].head(20) df['ERC20'].fillna(0) print(df['ERC20'].isna().sum()) df['whitelist']=df['whitelist'].map({'Yes': 1, 'No': 0}) print(df['whitelist'].head())

df['whitelist'].fillna(0) df['whitelist'].isna().sum() df['whitelist'].head()

#df['kyc'].head() df['whxky']=df['whitelist'] * df['kyc'] df['whxky'].head()

df['logsoldtokens']=np.log(df['sold_tokens']) print(df['logsoldtokens'].head()) df['logsoldtokens'] = df['logsoldtokens'].fillna(0) print(df['logsoldtokens'].head()) #for i in df['linkedin_link']: # if i = 'TBD': df.loc[df['linkedin_link']=="TBD", 'linkedin_link'] = 0 df['linkedin_link'].fillna(0) df.loc[df['linkedin_link']!=0, "linkedin_link"] = 1 #df['linkedin_link']=df['linkedin_link'].map({'TBD': 0, 'No': 0}) print(df['linkedin_link'].sum())

df['link_white_paper'].fillna(0, inplace=True) df.loc[df['link_white_paper']!=0, "link_white_paper"] = 1 print(df['link_white_paper'].sum()) print(df['link_white_paper'].head()) df['linkxwhitep']=df['linkedin_link'] * df['link_white_paper'] print(df['linkxwhitep'].head())

df['github_link'].fillna(0, inplace=True) df.loc[df['github_link']=="None", "github_link"] = 0 df.loc[df['github_link']!=0, "github_link"] = 1 print(df['github_link'].sum())

df['ico_start'][0]

df['ico_start'].fillna(0, inplace=True) print(df['ico_start'].head()) df['ico_end'].fillna(0, inplace=True) print(df['ico_end'].head())

for i in range(len(df)): if df['ico_end'][i] == 0 and df['ico_start'][i] != 0: df['ico_end'][i] = "10/24/2021"

df['ico_start']=pd.to_datetime(df['ico_start'], errors='coerce',dayfirst=False) df['ico_end']=pd.to_datetime(df['ico_end'], errors='coerce',dayfirst=False)

df['ico_length']=df['ico_end']-df['ico_start'] df['ico_length']=df['ico_length'].dt.days print(df['ico_length'].head())

for i in range(len(df)): if df['ico_length'][i] < 0: df['ico_length'][i] = 1

!pip install statsmodels

import statsmodels.api as sm from statsmodels.formula.api import ols, logit # dependent/target/outcome variable y = df['success'] # independent/predictor/explanatory variable X = df[['teamsize', 'rating']] # A. Logit regression # turn independent variables into floating type (best practice) # 'missing='drop'' drops rows with missing values from the regression logit_model=sm.Logit(y,X.astype(float), missing='drop' ) # fit logit model into the data result=logit_model.fit() # summarize the logit model print(result.summary2()) print(result.tvalues)

sns.regplot(x = "rating", y = "success", data = df, logistic = True, y_jitter = .05) plt.ylabel("success probability")

''' options for "at" 1. 'overall' The average of the marginal effects at each observation 2. 'mean' The marginal effects at the mean of each regressor 3. 'median' The marginal effects at the median of each regressor 4. 'zero' The marginal effects at zero for each regressor 5. 'all' The marginal effects at each observation. options for "method" 1. 'dydx' No transformation is made and amrginal effects are returned 2. 'eyex' estimate elasticities of variables in exog 3. 'dyex' estimate semi-elasticity 4. 'eydx' estimate semi-elasticity ''' average_marginal_effect = result.get_margeff(at = "mean", method = "dydx") print(average_marginal_effect.summary())

# B. Linear Probability Model # logit regression X = sm.add_constant(X) ols_model=sm.OLS(y,X.astype(float), missing='drop') result=ols_model.fit() print(result.summary2()) #print(result.rsquared_adj)

# Use wrapper lazypredict !pip install lazypredict

from lazypredict.Supervised import LazyClassifier, LazyRegressor from sklearn.model_selection import train_test_split

# load data X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=42) # fit all models clf = LazyClassifier(predictions=True) models, predictions = clf.fit(X_train, X_test, y_train, y_test)

models

predictions

models['Accuracy']['LogisticRegression']

def prsq(variables, target, basetable): X = basetable[variables] y = basetable[target] logreg = sm.Logit(y,X.astype(float), missing = 'drop') result=logreg.fit() #return(abs(result.tvalues)) return(result.prsquared)

def next_best(current_variables, candidate_variables, target, basetable): best_rsq = -1 best_variable = None for v in candidate_variables: logrsq = prsq(current_variables+[v], target, basetable) if logrsq >= best_rsq: best_rsq = logrsq best_variable = v return best_variable candidate_variables = ['bonus','sold_tokens','bounty','linkedin_link'] current_variables = ['teamsize','rating'] target = ['success'] next_variable = next_best(current_variables, candidate_variables,target, df) print(next_variable)

candidate_variables = ['teamsize','rating','bonus','sold_tokens','bounty','kyc'] #candidate_variables = ['teamsize','rating','bonus','bounty','kyc'] current_variables = [] target = ['success'] max_number_variables = 5 number_iterations = min(max_number_variables, len(candidate_variables)) for i in range(0,number_iterations): next_var = next_best(current_variables, candidate_variables, target, df) current_variables = current_variables + [next_var] candidate_variables.remove(next_var) print(current_variables)

# dependent/target/outcome variable y = df['success'] # independent/predictor/explanatory variable X = df[current_variables] # A. Logit regression # turn independent variables into floating type (best practice) # 'missing='drop'' drops rows with missing values from the regression logit_model=sm.Logit(y,X.astype(float), missing='drop' ) # fit logit model into the data result=logit_model.fit() # summarize the logit model print(result.summary2())

# B. Linear Probability Model # logit regression X = sm.add_constant(X) ols_model=sm.OLS(y,X.astype(float), missing='drop') result=ols_model.fit() print(result.summary2()) #print(result.rsquared_adj)

# load data X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=42) # fit all models clf = LazyClassifier(predictions=True) models, predictions = clf.fit(X_train, X_test, y_train, y_test)

print(models['Accuracy']['LogisticRegression']) print(models['ROC AUC']['LogisticRegression'])

def acctest(variables, target, basetable): X = basetable[variables] y = basetable[target] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=42) clf = LazyClassifier(predictions=True) models, predictions = clf.fit(X_train, X_test, y_train, y_test) #logreg = sm.Logit(y,X.astype(float), missing = 'drop') #result=logreg.fit() #return(abs(result.tvalues)) return(models['ROC AUC']['LogisticRegression'])

def next_best2(current_variables, candidate_variables, target, basetable): best_rsq = -1 best_variable = None for v in candidate_variables: logrsq = acctest(current_variables+[v], target, basetable) if logrsq >= best_rsq: best_rsq = logrsq best_variable = v return best_variable candidate_variables = ['bonus','sold_tokens','bounty'] current_variables = ['teamsize','rating'] target = ['success'] next_variable = next_best2(current_variables, candidate_variables,target, df) print(next_variable)

print(next_variable)

candidate_variables = ['teamsize','rating','bonus','sold_tokens','bounty','kyc'] current_variables = [] target = ['success'] max_number_variables = 5 number_iterations = min(max_number_variables, len(candidate_variables)) for i in range(0,number_iterations): next_var = next_best2(current_variables, candidate_variables, target, df) current_variables = current_variables + [next_var] candidate_variables.remove(next_var) print(current_variables)

# dependent/target/outcome variable y = df['success'] # independent/predictor/explanatory variable X = df[current_variables] # A. Logit regression # turn independent variables into floating type (best practice) # 'missing='drop'' drops rows with missing values from the regression logit_model=sm.Logit(y,X.astype(float), missing='drop' ) # fit logit model into the data result=logit_model.fit() # summarize the logit model print(result.summary2())

# load data X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=42) # fit all models clf = LazyClassifier(predictions=True) models, predictions = clf.fit(X_train, X_test, y_train, y_test)

print(models['Accuracy']['LogisticRegression']) print(models['ROC AUC']['LogisticRegression'])

df.columns[2:]

df['name'].isnull().sum()

#candidate_variables = ['teamsize','rating','bonus','logsoldtokens','bounty','kyc','ratxteamsize','is_ico','ERC20','whitelist','whxky','linkedin_link','link_white_paper','linkxwhitep','github_link','ico_length','raised_usd'] candidate_variables = ['teamsize','rating','bonus','logsoldtokens','bounty','kyc','ratxteamsize','is_ico','ERC20','whitelist','whxky','link_white_paper','linkxwhitep','github_link','ico_length'] current_variables = [] target = ['success'] max_number_variables = 8 number_iterations = min(max_number_variables, len(candidate_variables)) for i in range(0,number_iterations): next_var = next_best(current_variables, candidate_variables, target, df) current_variables = current_variables + [next_var] candidate_variables.remove(next_var) print(next_var) print(current_variables)

#candidate_variables = ['teamsize','rating','bonus','sold_tokens','bounty','kyc','ratxteamsize'] candidate_variables = ['teamsize','rating','bonus','logsoldtokens','bounty','kyc','ratxteamsize','is_ico','ERC20','whitelist','whxky','linkedin_link'] current_variables = [] target = ['success'] max_number_variables = 8 number_iterations = min(max_number_variables, len(candidate_variables)) for i in range(0,number_iterations): next_var = next_best2(current_variables, candidate_variables, target, df) current_variables = current_variables + [next_var] candidate_variables.remove(next_var) print(current_variables)

# dependent/target/outcome variable y = df['success'] # independent/predictor/explanatory variable #test_variables = ['teamsize','rating','bonus','logsoldtokens','bounty','kyc','ratxteamsize','is_ico','ERC20','whitelist','whxky','linkedin_link','link_white_paper'] test_variables = ['teamsize','rating','bonus','logsoldtokens','bounty','kyc','ratxteamsize','is_ico','ERC20','whitelist','whxky','linkedin_link','link_white_paper','linkxwhitep','github_link','ico_length'] #X = df[current_variables] X = df[test_variables] # load data X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=42) # fit all models clf = LazyClassifier(predictions=True) models, predictions = clf.fit(X_train, X_test, y_train, y_test) from sklearn.ensemble import RandomForestClassifier #Create a Gaussian Classifier clf=RandomForestClassifier(n_estimators=100) X_train.replace([np.inf, -np.inf], np.nan, inplace=True) X_train.fillna(0, inplace=True) X_train = np.nan_to_num(X_train.astype(np.float32)) clf.fit(X_train,y_train)

feature_imp = pd.Series(clf.feature_importances_,index=df[test_variables].columns).sort_values(ascending=False) feature_imp

import matplotlib.pyplot as plt import seaborn as sns %matplotlib inline # Creating a bar plot sns.barplot(x=feature_imp, y=feature_imp.index) # Add labels to your graph plt.xlabel('Feature Importance Score') plt.ylabel('Features') plt.title("Visualizing Important Features") plt.legend() plt.show()

#chosen_variables = ['raised_usd','is_ico','rating','teamsize','ratxteamsize','linkedin_link'] chosen_variables = ['is_ico', 'ratxteamsize', 'rating', 'bonus', 'teamsize', 'kyc','whitelist','logsoldtokens'] #chosen_variables = ['is_ico', 'ratxteamsize', 'bonus', 'teamsize', 'logsoldtokens', 'rating', 'linkxwhitep','linkedin_link','link_white_paper'] # dependent/target/outcome variable y = df['success'] # independent/predictor/explanatory variable X = df[chosen_variables] #X = df[current_variables] # A. Logit regression # turn independent variables into floating type (best practice) # 'missing='drop'' drops rows with missing values from the regression logit_model=sm.Logit(y,X.astype(float), missing='drop' ) # fit logit model into the data result=logit_model.fit() # summarize the logit model print(result.summary2())

# B. Linear Probability Model # logit regression X = sm.add_constant(X) ols_model=sm.OLS(y,X.astype(float), missing='drop') result=ols_model.fit() print(result.summary2()) #print(result.rsquared_adj)

# load data X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=.2, random_state=42) # fit all models clf = LazyClassifier(predictions=True) models, predictions = clf.fit(X_train, X_test, y_train, y_test) print('The accuracy of our model is',models['Accuracy']['LogisticRegression']) print('The ROC AUC of our model is',models['ROC AUC']['LogisticRegression'])

models