abhijeet-talaulikar

def hill_transform(x, alpha, gamma):
    return 1 / (1 + (x/gamma)**-alpha)

abhijeet-talaulikar

media = pd.read_csv('media.csv', parse_dates=['DATE'])
media = media.resample('W-Mon', label='left', closed='left', on='DATE').sum()

abhijeet-talaulikar

channel_priors = {
    "CTV": 0.02,
    "DISPLAY": 0.05,
    "DIRECT_MAIL": 0.02,
    "EMAIL": 0.03,
    "PAID_SOCIAL": 0.08,
    "PAID_SEARCH": 0.14,
    "TV": 0.01
}

abhijeet-talaulikar

# define coefficient
channel_prior = channel_priors[channel]
channel_coefficient = pm.TruncatedNormal(f"coefficient_{channel}", mu=channel_prior, sigma=0.0001, lower=0, upper=0.15)

# define saturation
alpha = pm.Uniform(f"alpha_{channel}", lower=0.5, upper=2)
gamma = pm.Uniform(f"gamma_{channel}", lower=0.5, upper=1.5)

abhijeet-talaulikar

def scale(x, y):
  return x * y.sum() / x.sum()

with pm.Model() as mmm:

    target = media_transformed['REVENUE'] / media_transformed['REVENUE'].mean()

    media_contributions = []

    for channel in channel_priors.keys():

abhijeet-talaulikar

trace_vars = [i.replace('coefficient_','') for i in dir(trace['posterior']) if i.startswith("coefficient_") or i=='intercept']
trace_coeff_vars = [i for i in dir(trace['posterior']) if i.startswith("coefficient_") or i=='intercept']
trace_alpha_vars = [i for i in dir(trace['posterior']) if i.startswith("alpha_")]
trace_gamma_vars = [i for i in dir(trace['posterior']) if i.startswith("gamma_")]

model_results = pd.DataFrame({
    "variable": trace_vars,
    "coefficient": [float(trace['posterior'][i].mean()) for i in trace_coeff_vars]
}).merge(
    pd.DataFrame({

abhijeet-talaulikar

az.plot_posterior(
    trace,
    var_names=['intercept', 'coefficient_'],
    filter_vars='like'
);

abhijeet-talaulikar

bias = pd.DataFrame({
    "t": np.arange(104)
})

d = 2

# Fourier terms
for i in np.arange(1,d+1):
    bias[f"cos_{i}"] = np.cos(2 * np.pi * i * bias["t"] / 52)
    bias[f"sin_{i}"] = np.sin(2 * np.pi * i * bias["t"] / 52)

abhijeet-talaulikar

media_decomp = pd.DataFrame({i:np.array(trace['posterior']["contribution_"+str(i)]).mean(axis=(0,1)) for i in channel_priors.keys()}, index=dates) * media['REVENUE'].mean()

media_roi = pd.concat([
    media.drop(['DATE','REVENUE'],axis=1).sum(), 
    media_decomp.sum()
], axis=1).reset_index()

media_roi.columns = ['media','spend','revenue']

media_roi['ROI'] = (media_roi['revenue'] / media_roi['spend'])

abhijeet-talaulikar

def get_response_curve(channel, start_time, end_time):
    
    def hill_transform(x, alpha, gamma):
        return 1 / (1 + (x/gamma)**-alpha)
    
    # parameters
    coefficient = model.query("variable == @channel")['coefficient'].iloc[0]
    alpha = model.query("variable == @channel")['alpha'].iloc[0]
    gamma = model.query("variable == @channel")['gamma'].iloc[0]